NO-Classifier: Prediction of Anti- Neuroinflammatory Agents Using Text Mining of 3D Molecular Fingerprints

NO-Classifier: Prediction of Anti- Neuroinflammatory Agents Using Text Mining of 3D Molecular Fingerprints | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article NO-Classifier: Prediction of Anti- Neuroinflammatory Agents Using Text Mining of 3D Molecular Fingerprints Si Eun Lee, Sangjin Ahn, Surendra Kumar, Mi-hyun Kim This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3812369/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 16 Nov, 2024 Read the published version in Scientific Reports → Version 1 posted 12 You are reading this latest preprint version Abstract CNS Drug discovery has been challenging due to the lack of clarity on CNS diseases' basic biological and pathological mechanisms. Despite the difficulty, some CNS drugs have been developed based on phenotypic effects. Herein, we propose a phenotype-structure relationship model, which predicts an anti-neuroinflammatory potency based on 3D molecular structures of the phenotype-active or inactive compounds without specifying targets. For this chemo-centric study, a predictive model of the nitric oxide (NO) inhibitory potency in hyper-activated microglia is built from the 548 agents, which were collected from 95 research articles (28 substructures consisting of natural products and synthetic scaffolds) and doubly externally validated by the agents of 9 research articles as third set. 3D Structures (multi-conformer ensemble) of every agent were encoded into the E3FP molecular fingerprint of the Keiser group as a 3D molecular representation. The location information of the molecular fingerprints could be learned and validated to classify the inhibitory potency of compounds (IC 50 cut-off between bi-classes: 37.1 μM): (1) multi-layer perceptron (MLP) (accuracy: 0.962, AUC: 0.994), (2) recurrent neural network (RNN) (accuracy: 0.966, AUC: 0.994), and (3) convolutional neural network (CNN) (accuracy: 0.969, AUC: 0.996). The high performance of these models was compared with that of four classical machine classification models (Logistic, Ridge, Lasso, and Naïve Bayes). We named the bi-class models NO-Classifier. Out-of-set validation and decision region analysis of the out-of-set doubly demonstrated NO-Classifier effectively discerned the anti-inflammatory potency of testing compounds in inflammatory cell phenotype with the webserver in https://no-classifier.onrender.com. Biological sciences/Computational biology and bioinformatics/Computational models Biological sciences/Computational biology and bioinformatics/Data processing Biological sciences/Computational biology and bioinformatics/High throughput screening Biological sciences/Computational biology and bioinformatics/Machine learning Biological sciences/Drug discovery/Drug screening Biological sciences/Computational biology and bioinformatics Biological sciences/Drug discovery Health sciences/Molecular medicine Physical sciences/Chemistry Physical sciences/Chemistry/Cheminformatics Molecular Featurization Phenotype Structure Relationship Neuroinflammation Microglia Activation Location Information Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Highlights NO-Classifier is a phenotype-structure relationship model NO-Classifier predicts an anti-neuroinflammatory potency of natural and non-natural compounds NO inhibitory potency in hyperactivated microglia was trained with the location information of 3D molecular structures The statistical performance of NO-Classifier with accuracy of 0.969 and AUC of 0.996 1. Introduction The lack of clarity in basic biological and pathological mechanisms related to brain has made CNS drug discovery difficult. Despite the difficulty, some CNS drugs have been developed based on phenotypic effects. Neuroinflammation is one of hallmark phenotypes observed in neurological disorders1, which include age-related dementia2, neurodegenerative diseases such as amyotrophic lateral sclerosis (ALS), Parkinson's disease (PD), and Alzheimer's disease (AD)3–5. Neuroinflammation has been well studied in cellular level. For example, lipopolysaccharide (LPS) can induce the inflammation of neuronal cells through the hyperactivation of microglia which plays proinflammatory roles in CNS6. Moreover, because LPS does not involve CNS disruption, it is used as a stimulant to generate neuroinflammation. The hyperactivated microglia generated inflammatory substances such as nitric oxide (NO). The NO level in cells contributes to the balance between pro-inflammation and anti-inflammation7. Thus, NO production in LPS induced microglial cells can be used for measuring neuroinflammation of cells and NO production assay is one of the simple screening methods for anti-neuroinflammatory agents. Because NO inhibitory potency of a testing compound observed in the assay is an effective indicator to show how much the testing compound can suppress neuroinflammation, we judged that the testing data of NO production assay can be used for a phenotype-structure relationship model. Surely, the model should be built using phenotype-active or inactive compounds without any information on specific targets. When the phenotype-structure relationship model is realized, the chemo-centric study can aid the drug design and drug screening for neuroinflammation-related diseases such as ALS, PD, and AD. This motivation made us developed a classification model to predict anti-inflammatory potency of various compounds of > 29 multiple scaffolds using their valid IC50 values of NO assay. Our sequential research studies on 3D chemical similarity and diverse molecular featurization for drug discovery encouraged to investigate molecular features for this phenotype model. In this study, we report the location information of 3D molecular structure as an efficient molecular feature and the 3D location information based classification model using neural network and conventional classifiers. 2. Methods 2.1 Dataset Selection and Preparation We collected academic papers including compounds related with neuroinflammation using Google Scholar and EMBASE. The criteria of the collection were structural diversity of the compounds and the balance between natural occurring and artificial synthesizing. Moreover, the artificial synthetic molecules were mainly in-house data accumulated in our serial studies for neuroinflammation 8,9 . While 95 papers of 122 papers allowed us to extract compounds structures and their biological activities through Reaxys and Pubchem. The collected compounds were sorted by their pharmacological effects: 584 compounds for inhibiting NO production, 197 compounds for inducing nerve growth factor, 5 compounds for promoting neurite outgrowth, and fewer compounds for other functions such as ameliorating neurotoxicity. Surely, the number of NO-inhibiting compounds was dominant, and thus we chose NO assay to describe drugs’ effect on neuroinflammation and used the corresponding data set ( Supplementary Table S1 -S2 ). When a compound has multiple IC50 values, the largest IC50 value was chosen among them. After removing 30 compounds owing to their severe toxicity, we gained the IC50 distribution of the dataset and decided the threshold (IC50 cut-off: 37 µM) values between active and inactive in the distribution (Fig. 1 ). In summary, 584 compounds of 65 papers were used for model training and validation 10–74. Moreover, another dataset was collected and prepared to use out-of-set 75–83. 2.2 3D-Molecular Representation and Featurization In order to generate 3D molecular representation, 3D conformers should be priori generated and sampled using Schrödinger Suite 84. We intended to calculate 3D molecular fingerprints85 from 3D conformers. Our data matrix comprising the SMILES of the collected compounds, their respective names, IC50 values, and other activities was manipulated using MAESTRO and Canvas of Schrödinger Suite86-88 and each SMILES was transformed to 3D conformation using LigPrep89. Multiple conformers were further generated using ConfGen90 , 91. Different numbers of rotatable bonds present in various compounds still made the multiple conformers cannot enough represent the conformational space of flexible compounds. Thus, conformer ensembles were more sampled based on the generated multiple conformers using Omega of Openeye 92 , 93 employing the default parameters. Finally, 60,275 3D conformers of 544 compounds were successfully acquired. The conformer ensemble of 60,275 size was employed to enhance diversity and scale of the dataset based on the literature 94, where conformational oversampling enhanced a dataset by generation of multiple conformations of a molecule and gained a desirable balance between sensitivity and specificity. Subsequently, E3FP of 1024-bit length as a 3D molecular fingerprint was generated from the conformer ensemble. E3FP encodes three-dimensional molecular geometry regardless of chemical bonds’ presence or absence ( https://github.com/keiserlab/e3fp)8 5. Upon computation of the 3D fingerprint, extant substructures are encoded to one (on-bit) within the fixed 1024-bits and some absence is transformed to zero (off-bit) in the array (as depicted in Fig. 2 ), thereby obtaining relative locational information of atoms95. While E3FP can be generally used as a sparse bitvector or bitstring, herein we tried to modify the sparse 1024 bit into the compact array of irregular length (the maximum length: 212). In detail, the location of every on-bit information was written to our positional array. And then we tokenized every element in the positional array using Keras Tokenizer96 and vectorized the relative location information of on-bit into an integer list using TF-IDF Vectorizer from scikit-learn 97. The above process applied to the third set as well as training/test set. Finally, we used the vectorized "positional information" as feature set for the two type bi-class classification models, which are neural networks using Keras and conventional machine learning using scikit-learn, and classification analysis to choose the best NO classifier model in Jupyter Notebook environment. 2.3 Model training As the above mentioned, Keras/TensorFlow and scikit-learn were chosen among the ideal Python libraries for machine learning96 , 97 in order to train NO inhibitory function of drugs and their 3D structures with reliability and efficacy. Based on three neural networks (NN) architectures and four classifier analysis (CA) methods, a total of seven discrete models were built. The construction and training of all models were executed in the condition of randomly chosen 80% of the dataset, the external validation was performed with the remaining 20% as test sets. 2.3.1 Neural Network We built three NN architectures of MLP, convolutional neural network (CNN), and recurrent neural network (RNN) to predict NO inhibitory potency of testing drugs (Fig. 2 ). The initial parameters for the NN architectures were tested with the epoch number of 15, batch size of 256, and learning rate of 0.001. The MLP and RNN have one dense layer with 32 units for each model, whereas the CNN has two dense layers. In detail, the MLP layer parameters (1024, 32, 212) define the vocabulary size, the dimensionality of the embedding space, and the maximum sequence length of the input text data, respectively. In other words, the maximum word index of the location information was 1024, which was the vocabulary size for word indexing from 0 to 1023. Each word was converted into a 32-dimensional vector. The maximal sequence length of the input drugs was 212 words. Subsequently, the flatten layer transforms the multidimensional input of shape (None, 212, 32) into a one-dimensional vector of shape (None, 6784), effectively collapsing all spatial dimensions. Lastly, the dense layer outputs a single value per batch. Similarly, after the same MLP, the RNN model incorporated a long short-term memory (LSTM) layer, which is designed for retaining information over extended sequences, effectively replacing the flatten layer. In addition to the aforementioned architectures, the CNN model incorporated a GlobalMaxPool1D layer for pooling operations, followed by two dense layers employing ReLU and sigmoid activations, respectively. All three NN models used binary cross-entropy as the loss function to calculate the gradients, and the RMSprop algorithm, an adaptive variant of the standard gradient descent, to optimize the parameters 98. Owing to the intrinsic stochasticity of neural network training, each model was preserved in the H5 file format. Under the condition, the variations in AUC values were contained within a narrow margin, not exceeding 0.1. 2.3.2 Classifier Analysis In this study, we tested the performance of several CA techniques, including logistic classification, ridge classification, least absolute shrinkage selection operator (LASSO), and Naïve Bayes. Evert model employed the TF-IDF Vectorizer to transform the location information into a matrix of inverse frequency features for each word. The logistic classifier was initialized with a random state of 0 as the seed value for the data shuffling process. Ridge Classifier was chosen based on the characteristics of the method, L2 penalty to address multicollinearity and feature selection to identify the most relevant variables for the classification task. Similarly, the LASSO classifier performed feature subset based on the convex optimization (L1 penalty)100 , 101. Naïve Bayes classifier is known to be an effective machine-learning method for text mining. The rapid and advanced method has been applied to various real-world scenarios, including spam filtering, document categorization, and text classification. Naïve Bayes classifier offers a range of models, including the Multinomial, Gaussian, and Bernoulli models, which Bernoulli and Gaussian event models implemented in scikit learn are not fully Bayesian 102. We developed a Gaussian Naïve Bayes (Gaussian NB) classifier to extract features, including word counts. 2.4 Model validation To develop the optimal predictive model, the built models was evaluated by several statistical metrics of Table 1 such as mean accuracy, specificity, sensitivity, F1-Score, area under the curve (AUC), and Matthews Correlation Coefficient (MCC), which is a particular case of the phi coefficient, ϕ, with a value between − 1 and 1103 , 104. Moreover, the confusion matrices of models were directly compared for the performance evaluation. The common statistical metrics for evaluating classification models were mathematically defined in the Table 1 . In this study, ACC, F1-score, AUC, MCC were mainly used to compare the performance between models. Obviously, ACC is conventional metric but cannot properly explain models built from imbalanced dataset between active and inactive. To compensate the limitation, F1-score and MCC were chosen and MCC was preferred for us. In general, MCC values more than 0.7 indicate the significant correlation between the predicted and actual classification. MCC values in the range of 0.4 to 0.7 for the MCC suggest a moderate level of the correlation, ranging from 0.4 to 0.7. Theoretically, while MCC of zero would result in a random prediction, MCC values ranging from 0 to 0.4 still have the correlation of low strength but superior to the random and MCC values below 0 indicate the inverse correlation104. (Table 1 ). Clearly, MCC is a contingency matrix technique used to determine the Pearson product-moment correlation coefficient. Therefore, it has the same interpretation as Pearson's r correlation105. Moreover, AUC was calculated from receiver operator characteristic (ROC) curves, which is plot of the true positive rate against the false positive rate. In detail, we used AUC values to judge discriminative potential of models: excellent (> 0.90), good (> 0.80), fair (> 0.70), and poor discrimination (< 0.70)106. Table 1 Evaluation metrics for the classification models. Metrics Equation Sensitivity (SE) \(\frac{TP}{TP+FN}\) Specificity (SP) \(\frac{TN}{FP+TN}\) Accuracy (ACC) \(\frac{TP+TN}{TP+TN+FP+FN}\) F1 score \(\frac{2TP}{2TP+FP+FN}\) MCC \(\frac{TP*TN-FP*FN}{\sqrt{\left(\left(TP+FP\right)*\left(TP+FN\right)*\left(TN+FP\right)*\left(TN+FN\right)\right)}}\) N: total number of dataset, TP: true positives, TN: true negatives, FN: false negatives, and FP: false positives. 3. Results and Discussion 3.1 Characterization of Anti-Inflammatory Dataset We collected anti-inflammatory agents and their activity information based on cell-based assay from academic articles10-74, Reaxys, EMBASE, and Pubchem. After the data regularization, we found NO production assay, which measure the intensity of pro-inflammation in hyperactivated microglia, has the massive data in the terms of scaffold diversity (Fig. 3 ). Therefore, the assay data as dataset was selected, and the IC50 value of NO production assay was used to describe anti-inflammatory potency of testing drugs. Before building model, the characterization of the dataset was performed through the scaffold analysis of compounds in the dataset and distribution of IC50 values. Under the IC50 cut-off of 37.1, the ratio between active and inactive was 1: 1. Moreover, 27 compounds of the same assay (LPS-induced NO assay) were further collected and used as the out-of-set to validate the models 75–83. The third set compounds were separated to csv files according to the scaffolds such as steroid, lignan, phenylpropanoid, iridoid, coumarin, and flavonoid. 3.2 Text Mining Based Molecular Featurization Herein, we intended to predict the molecular modulation of neuroinflammation by the three NN and four CA models. We named the models NO-Classifier. In detail, the primary aim of the built models is to explain the correlation of the 3D structures of testing compounds with their NO inhibitory potency and to predict potency of new compounds using their 3D structures. The final aim is to design new competent modulators for neuroinflammation treatment using the predictive models. For the purpose, E3FP location information was chosen for molecular featurization of the 3D structures (exactly, 3D molecular geometry). In detail, 3D Molecular representation (E3FP of 1024-bit length) was generated from conformer ensemble ( N = 60,275 of 544 compounds), which was prepared through conformational oversampling to enhance diversity and scale of the dataset based on the literature 94. It is well-known that one 2D molecular fingerprint can be decoded to one unique 2D molecule but 3D molecular fingerprint cannot be perfectly decoded to one 3D conformation85. Moreover, how to use enormous 3D conformational space has been one huddle in drug design and screening 107,108 .Thus, although E3FP, itself has been directly used as molecular descriptor, we intended to compress the information in E3FP for the cost effectiveness of the 3D data. At least, this trial reduced the absolute size of the 3D data. The transformation of sparse E3FP bitvector was conducted through gathering positional information of only present substructures (on-bit). As the result, 3D molecular geometry of each conformer was written to the dense array (of Fig. 4 ) like text data of irregular length. The relative location information of atom level substructure was applied to natural language processing (NLP) tools. In other words, we treated each value in the dense array as each word in a sentence. Because the conformers of 60,275 presented the maximum length of 212, the maximum length is in the range of general sentences for NLP. Thus, we tested some NLP tools such as NLTK (Natural Language Toolkit), Keras Processing, Word2vec, and Doc2vec. Conventional NLP follows the process consisting of tokenization, vectorization, and so on for feature MLP. We also tokenized every element of the positional array using Keras Tokenizer 96 and vectorized the relative location information of on-bit into the transformed inverse frequency feature using TF-IDF Vectorizer as a probability weighted information, 97,109. Finally, we used the vectorized "positional information" as feature set for our classification models. 3.3 Performance Evaluation of NO-Classifier NN and CA models were built from the NLP treated feature of positional information and their predictive power for the molecular modulation of neuroinflammation was evaluated by several statistical metrics as shown in the method section (2.4 Model Validation). The Tables 2 describes the internal 10-fold validation results using training data (80% of the dataset) and external validation using test data (20% of of the dataset) under the random dividing condition with specific seed numbers. Every NNs showed superior performance to CA models in the validation. Particularly, CNN demonstrated the best performance in terms of every statistical metric (Table 2 ). Table 2 Prediction performance of classification models ML Method External Validation 10-Fold Internal Cross-Validation ACC SP SE F1 Score AUC MCC ACC SP SE F1 Score AUC MCC MLP 0.962 0.968 0.952 0.952 0.994 0.920 0.983 0.984 0.981 0.979 0.999 0.965 RNN 0.966 0.973 0.955 0.957 0.994 0.928 0.982 0.985 0.976 0.977 0.999 0.962 CNN 0.969 0.973 0.961 0.961 0.996 0.935 0.983 0.984 0.982 0.980 0.999 0.966 Logistic 0.886 0.899 0.865 0.855 0.951 0.761 0.892 0.902 0.876 0.865 0.958 0.775 Ridge 0.884 0.897 0.864 0.853 0.948 0.758 0.887 0.899 0.869 0.859 0.954 0.764 LASSO 0.885 0.899 0.865 0.854 0.950 0.759 0.890 0.901 0.874 0.864 0.956 0.772 Naive Bayes 0.779 0.844 0.700 0.739 0.861 0.552 0.840 0.956 0.716 0.812 0.865 0.696 ACC: accuracy, SP: specificity, SE: sensitivity, F1-score: see equation in Table 2 , AUC: area under curve, MCC: Matthews’s correlation coefficient Among logistic classifier, ridge classifier, LASSO classifier, and Naïve Bayes classifier, LASSO was the best CA model but it’s statistical metrics was slightly lower than those of NN models. Notably, the AUC values of CA models satisfied the excellent criterion (> 0.90) with more than 0.95 except for Naïve Bayes classifier (AUC: 0.861), which still can be considered a commendable score106. Despite other statistical metrics inferior to NN models, every CA model presented the significant correlation between the predicted and actual classification based on the MCC values of > 0.7 except for Naïve Bayes classifier (moderate significant correlation with MCC of 0.696 for 10-fold cross validation and 0.552 for external validation)104. In summary, the findings suggest that our NLP treated 3D molecular feature works properly to judge the anti-inflammatory potency within chemical space and conformational space of the data set. The results encouraged us to further investigate the proficiency of our NO-Classifier in identifying new molecular modulators for neuroinflammation. 3.4 Out-of-Set Validation of NO-Classifier Further investigation of NO-Classifier was designed using ‘out-of-set’ data, which were not included in the dataset for either model training or validation. It is also called 3rd set. The new data can be grouped into two categories according to their sub-structures (scaffolds). If some new compound is absent in either training or test data but the scaffold of the compound is in the training or test data, the compound is included in the first category. Meanwhile, if both a new compound and its scaffold are absent in the training or test data, the compound is included in the second category. We expected the predictive performance of NO-Classifier to be retained for compounds in the first category. Meanwhile, we forecast that decreased performance for new scaffolds and wondered whether the second category compounds were compatible for NO-Classifier or not. As our expectation, the coumarin and flavonoid compounds (exact expression: derivatives) in the first category demonstrated promising outcomes concerning the AUC of the CNN (Fig. 5 ), with values of 0.913 (for coumarin) and 0.947 (for flavonoids). The compounds with saponin scaffold, which is absent in either the training data or test data, gave us the AUC value decreased to 0.760 in the CNN model. When we consider the dissimilarity of saponin derivatives with the dataset, the fair (> 0.70) AUC results of CNN model demonstrates the prediction potential of NO-Classifier applicable for future new coming compounds. In sequence, the out-of-set validation was performed for the CA models. In the case of flavonoid derivatives, the AUC values of logistic classifier (0.892) and LASSO classifier (0.867) exhibited good results (> 0.80) and Naïve Bayes classifier gave fair AUC of 0.718. Despite the first category of coumarins, CA models didn’t present fair AUC values. It suggests other factors except for common scaffolds. When flavonoid is compared with coumarin, the conformational rigidity could generate the limited number of conformers. Thus, we guess the rigidity of common scaffolds can make an effect on prediction performance. Our models tended to work more properly for compounds with a flexible scaffold able to have the high coverage for conformational space to predict future upcoming compounds. Remarkably, saponin, which was not incorporated into the training-test set and exhibited low similarity, demonstrated divergent outcomes in both CNN model and classical analysis models. To doubly confirm the performance of NO-Classifier, we produced decision regions (or decision boundary) for the CA models. The analysis with visualized maps revealed that the CA models exhibited a discerning ability towards flavonoid derivatives. Interestingly, despite the structural dissimilarity of saponin, decision regions also demonstrated effectively discerning performance. Once more, it is worth noting that almost compounds, which were gathered explicitly, fell into their right classification (Fig. 6 ). Thus, the classification models and decision regions demonstrated reliable performance in effectively discerning the anti-inflammatory potency of compounds in inflammatory cell phenotype. 4. Conclusion We developed NO-Classifier using three NN and four CA algorithms, which is classification models for predicting anti-inflammatory potency against neuroinflammation. The models learned the 3D location information of multiple structures (conformer ensembles of drugs) as molecular feature and IC50 values of NO assay. NO-Classifier satisfied statistical metrics for both internal and external validation to exhibit the classification performance. Moreover, the satisfactory performance across all assessed metrics was doubly confirmed by both out-of-set validation and decision region analysis of the out-of-set. It demonstrates NO-Classifier can be used for forecasting anti-inflammatory potency of new coming compounds. At least, NO-Classifier can screen drugs to suppress neuroinflammatory process in neurodegenerative diseases. In the future, the prediction coverage of NO-Classifier can be expanded to more diverse and different compounds using the updated molecular data. Declarations Data Availability & Source Codes Supplementary file is available. All source code, and data are available in GitHub. https://github.com/college-of-pharmacy-gachon-university/NO-Classifier The webserver is available: https://no-classifier.onrender.com Acknowledgement & Funding The authors would like to express their gratitude to Professor Sun Yeou Kim and her laboratory for their invaluable suggestion and contribution to the initial data collection. The authors would like to thank OpenEye Scientific Software for providing an academic free license. This study was supported by the Basic Science Research Program of the National Research Foundation of Korea (NRF), which is funded by the Ministry of Education, Science and Technology (No.: 2022R1A2C2091810). Authors’ Contribution M.K. conceived and designed the study. Under M. K.’s plan, J.A. and S.L. investigated diverse molecular featurization and S.L. fully conducted data collection and manipulation. M. K., and S. L. analyzed every data and result. S.K. refined source codes and developed webserver. M. K., and S. L. wrote the manuscript and revised it. M.K. provided the molecular modeling lab facility. All authors read and approved the final manuscript. The authors confirm that this article content has no conflicts of interest. References Schwartz, M.; Deczkowska, A., Neurological disease as a failure of brain–immune crosstalk: the multiple faces of neuroinflammation. Trends in immunology 2016, 37, 668–679. Lynch, M. A., Age-related neuroinflammatory changes negatively impact on neuronal function. Frontiers in Aging Neuroscience 2010, 1, 6. Heneka, M. T.; O'Banion, M. K., Inflammatory processes in Alzheimer's disease. Journal of Neuroimmunology 2007, 184, 69–91. Frank-Cannon, T. C.; Alto, L. T.; McAlpine, F. E.; Tansey, M. G., Does neuroinflammation fan the flame in neurodegenerative diseases? Molecular neurodegeneration 2009, 4, 1–13. Heppner, F. L.; Ransohoff, R. M.; Becher, B., Immune attack: the role of inflammation in Alzheimer disease. Nature Reviews Neuroscience 2015, 16, 358–372. Chen, Z.; Jalabi, W.; Shpargel, K. B.; Farabaugh, K. T.; Dutta, R.; Yin, X.; Kidd, G. J.; Bergmann, C. C.; Stohlman, S. A.; Trapp, B. D., Lipopolysaccharide-induced microglial activation and neuroprotection against experimental brain injury is independent of hematogenous TLR4. Journal of Neuroscience 2012, 32, 11706–11715. Bourgognon, J.-M.; Spiers, J. G.; Robinson, S. W.; Scheiblich, H.; Glynn, P.; Ortori, C.; Bradley, S. J.; Tobin, A. B.; Steinert, J. R., Inhibition of neuroinflammatory nitric oxide signaling suppresses glycation and prevents neuronal dysfunction in mouse prion disease. Proceedings of the National Academy of Sciences 2021, 118. Venkanna, A.; Kwon, O. W.; Afzal, S.; Jang, C.; Cho, K.; Yadav, D. K.; Kim, K.; Park, H.-g.; Chun, K.-H.; Kim, S. Y.; Kim, M.-h., Pharmacological use of a novel scaffold, anomeric N,N-diarylamino tetrahydropyran: molecular similarity search, chemocentric target profiling, and experimental evidence, Scientific Reports, 2017, 7, 12535 Venkanna, A.; Subedi, L.; Teli, M. K.; Lama, P. D.; Nangunuri, B. G.; Lee, S. Y.; Kim, S.Y.; Kim, M. H., Positioning of an unprecedented spiro [5.5] undeca ring system into kinase inhibitor space. Scientific Reports, 2020, 10(1), 21265. Kim, D. H.; Kim, C. S.; Subedi, L.; Kim, S. Y.; Lee, K. R., Alkaloids of NIRAM, natural dye from Polygonum tinctorium, and their anti-inflammatory activities. Tetrahedron Letters 2019, 60, 151130. Kim, K. H.; Kim, C. S.; Park, Y. J.; Moon, E.; Choi, S. U.; Lee, J. H.; Kim, S. Y.; Lee, K. R., Anti-inflammatory and antitumor phenylpropanoid sucrosides from the seeds of Raphanus sativus. Bioorganic Medicinal Chemistry Letters 2015, 25, 96–99. Subedi, L.; Lee, J. H.; Yumnam, S.; Ji, E.; Kim, S. Y., Anti-inflammatory effect of sulforaphane on LPS-activated microglia potentially through JNK/AP-1/NF-κB inhibition and Nrf2/HO-1 activation. Cells 2019, 8, 194. Park, K. J.; Subedi, L.; Kim, S. Y.; Lee, K. R., Anti-inflammatory terpenoid derivatives from the twigs of Syringa oblata var. dilatata. Phytochemistry Letters 2018, 27, 183–186. Kim, C. S.; Bae, M.; Oh, J.; Subedi, L.; Suh, W. S.; Choi, S. Z.; Son, M. W.; Kim, S. Y.; Choi, S. U.; Oh, D.-C., Anti-Neurodegenerative Biflavonoid Glycosides from Impatiens balsamina. Journal of Natural Products 2017, 80, 471–478. Ha, S. K.; Shobha, D.; Moon, E.; Chari, M. A.; Mukkanti, K.; Kim, S.-H.; Ahn, K.-H.; Kim, S. Y., Anti-neuroinflammatory activity of 1,5-benzodiazepine derivatives. Bioorganic Medicinal Chemistry Letters 2010, 20, 3969–3971. Lee, S. Y.; Suh, W. S.; Cha, J. M.; Moon, E.; Ha, S. K.; Kim, S. Y.; Lee, K. R., Anti-neuroinflammatory constituents from Sinomenium acutum rhizomes. Phytochemistry Letters 2016, 17, 79–84. Kim, K. H.; Lee, K. H.; Kim, H. K.; Moon, E.; Kim, S.-H.; Kim, S. Y.; Kim, K. R.; Lee, K. R., Antineuroinflammatory constituents from the root extract of Paris verticillata. Canadian Journal of Chemistry 2011, 89, 441–445. Woo, K. W.; Moon, E.; Kwon, O. W.; Lee, S. O.; Kim, S. Y.; Choi, S. Z.; Son, M. W.; Lee, K. R., Anti-neuroinflammatory diarylheptanoids from the rhizomes of Dioscorea nipponica. Bioorganic Medicinal Chemistry Letters 2013, 23, 3806–3809. Suh, W. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Bioactive lignan constituents from the twigs of Sambucus williamsii. Bioorganic Medicinal Chemistry Letters 2016, 26, 1877–1880. Woo, K. W.; Suh, W. S.; Subedi, L.; Kim, S. Y.; Kim, A.; Lee, K. R., Bioactive lignan derivatives from the stems of Firmiana simplex. Bioorganic Medicinal Chemistry Letters 2016, 26, 730–733. Kim, K. H.; Kim, H. K.; Choi, S. U.; Moon, E.; Kim, S. Y.; Lee, K. R., Bioactive lignans from the rhizomes of Acorus gramineus. Journal of Natural Products 2011, 74, 2187–2192. Kim, C. S.; Kwon, O. W.; Kim, S. Y.; Lee, K. R., Bioactive lignans from the trunk of Abies holophylla. Journal of Natural Products 2013, 76, 2131–2135. Kim, K. H.; Ha, S. K.; Choi, S. U.; Kim, S. Y.; Lee, K. R., Bioactive phenolic constituents from the seeds of Pharbitis nil. Chemical Pharmaceutical Bulletin 2011, 59, 1425–1429. Kim, K. H.; Moon, E.; Kim, S. Y.; Choi, S. U.; Son, M. W.; Choi, S. Z.; Lee, K. R., Bioactive sesquiterpenes from the essential oil of Thuja orientalis. Planta Medica 2013, 79, 1680–1684. Kim, C. S.; Subedi, L.; Oh, J.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Bioactive Triterpenoids from the Twigs of Chaenomeles sinensis. Journal of Natural Products 2017, 80, 1134–1140. Park, K. J.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Bioactive triterpenoids from twigs of Betula schmidtii. Bioorganic Chemistry 2018, 77, 527–533. Kim, K. H.; Moon, E.; Choi, S. U.; Kim, S. Y.; Lee, K. R., Biological evaluation of phenolic constituents from the trunk of Berberis koreana. Bioorganic Medicinal Chemistry Letters 2011, 21, 2270–2273. Suh, W. S.; Park, K. J.; Kim, D. H.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., A biphenyl derivative from the twigs of Chaenomeles speciosa. Bioorganic Chemistry 2017, 72, 156–160. Kim, K. H.; Choi, S. U.; Ha, S. K.; Kim, S. Y.; Lee, K. R., Biphenyls from Berberis koreana. Journal of Natural Products 2009, 72, 2061–2064. Kim, D. H.; Lee, T. H.; Subedi, L.; Kim, S. Y.; Lee, K. R., Chemical constituents of Impatiens balsamina stems and their biological activities. Natural Product Sciences 2019, 25, 130–135. Kim, K. H.; Choi, J. W.; Choi, S. U.; Ha, S. K.; Kim, S. Y.; Park, H.-J.; Lee, K. R., The chemical constituents of Piper kadsura and their cytotoxic and anti-neuroinflammtaory activities. J Enzym Inhib Med Ch 2011, 26, 254–260. So, H. M.; Yu, J. S.; Khan, Z.; Subedi, L.; Ko, Y.-J.; Lee, I. K.; Park, W. S.; Chung, S. J.; Ahn, M.-J.; Kim, S. Y., Chemical constituents of the root bark of Ulmus davidiana var. japonica and their potential biological activities. Bioorganic Chemistry 2019, 91, 103145. Kim, K. H.; Moon, E.; Lee, S. R.; Park, K. J.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Chemical Constituents of the Seeds of Raphanus sativus and their Biological Activity. Journal of the Brazilian Chemical Society 2015, 26, 2307–2312. Venkanna, A.; Cho, K. H.; Dhorma, L. P.; Kumar, D. N.; Hah, J. M.; Park, H.-g.; Kim, S. Y.; Kim, M.-h., Chemistry-oriented synthesis (ChOS) and target deconvolution on neuroprotective effect of a novel scaffold, oxaza spiroquinone. European Journal of Medicinal Chemistry 2019, 163, 453–480. Hyun Kim, K.; Keun Ha, S.; Yeou Kim, S.; Joo Youn, H.; Ro Lee, K., Constituents of Limonia acidissima inhibit LPS-induced nitric oxide production in BV-2 microglia. Journal of enzyme inhibition medicinal chemistry 2010, 25, 887–892. Kim, C. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Kim, K. H.; Lee, K. R., Diterpenes from the Trunk of Abies holophylla and Their Potential Neuroprotective and Anti-inflammatory Activities. Journal of Natural Products 2016, 79, 387–394. Woo, K. W.; Moon, E.; Park, S. Y.; Kim, S. Y.; Lee, K. R., Flavonoid glycosides from the leaves of Allium victorialis var. platyphyllum and their anti-neuroinflammatory effects. Bioorg Med Chem Lett 2012, 22, 7465–7470. Kim, C. S.; Oh, J.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Holophyllane A: A Triterpenoid Possessing an Unprecedented B-nor-3,4-seco-17,14-friedo-lanostane Architecture from Abies holophylla. Scientific Reports 2017, 7, 1–8. Kim, C. S.; Shin, B.; Kwon, O. W.; Kim, S. Y.; Choi, S. U.; Oh, D.-C.; Kim, K. H.; Lee, K. R., Holophyllin A, a rearranged abietane-type diterpenoid from the trunk of Abies holophylla. Tetrahedron Letters 2014, 55, 6504–6507. Kim, C. S.; Oh, J.; Suh, W. S.; Jang, S. W.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Investigation of chemical constituents from Spiraea prunifolia var. simpliciflora and their biological activities. Phytochemistry Letters 2017, 22, 255–260. Suh, W. S.; Kim, C. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Iridoid Glycosides from the Twigs of Sambucus williamsii var. coreana and Their Biological Activities. Journal of Natural Products 2017, 80, 2502–2508. Kim, C. S.; Subedi, L.; Kwon, O. K.; Kim, S. Y.; Yeo, E.-J.; Choi, S. U.; Lee, K. R., Isolation of bioactive biphenyl compounds from the twigs of Chaenomeles sinensis. Bioorganic Medicinal Chemistry Letters 2016, 26, 351–354. Kim, K. H.; Moon, E.; Choi, S. U.; Kim, S. Y.; Lee, K. R., Lanostane triterpenoids from the mushroom Naematoloma fasciculare. Journal of Natural Products 2013, 76, 845–851. Kim, C. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Kim, K. H.; Lee, K. R., Lignan Glycosides from the Twigs of Chaenomeles sinensis and Their Biological Activities. Journal of Natural Products 2015, 78, 1174–1178. Kim, K.-H.; Ha, S.-K.; Kim, S.-Y.; Kim, S.-H.; Lee, K.-R., Limodissimin A: A New Dimeric Coumarin from Limonia acidissima. Bulletin of the Korean Chemical Society 2009, 30, 2135–2137. Dung, H. V.; Bach, N. V.; Trung, T. N.; Nhiem, N. X.; Tai, B. H.; Kiem, P. V.; Park, S.; Lee, T. H.; Kim, S. Y.; Kim, S. H., Megastigmane glycosides from Docynia indica and their anti-inflammatory activities. Helvetica Chimica Acta 2016, 99, 681–686. Subedi, L.; Kwon, O. W.; Pak, C.; Lee, G.; Lee, K.; Kim, H.; Kim, S. Y., N,N-disubstituted azines attenuate LPS-mediated neuroinflammation in microglia and neuronal apoptosis via inhibiting MAPK signaling pathways. BMC Neuroscience 2017, 18, 1–12. Kim, K. H.; Choi, J. W.; Ha, S. K.; Kim, S. Y.; Lee, K. R., Neolignans from Piper kadsura and their anti-neuroinflammatory activity. Bioorganic medicinal chemistry letters 2010, 20, 409–412. Kim, C. S.; Suh, W. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Neuroprotective Fatty Acids from the Stem Bark of Sorbus commixta. Lipids 2016, 51, 989–995. Van Kiem, P.; Cuong, L. C. V.; Trang, D. T.; Nhiem, N. X.; Anh, H. L. T.; Tai, B. H.; Huong, L. M.; Van Minh, C.; Lee, T. H.; Kim, S. Y., New Alkaloids and Anti-inflammatory Constituents from the Leaves of Antidesma ghaesembilla. Natural Product Communications 2017, 12, 11–14. Lee, T. H.; Khan, Z.; Subedi, L.; Kim, S. Y.; Lee, K. R., New bis-thioglycosyl-1, 1′-disulfides from Nasturtium officinale R. Br. and their anti-neuroinflammatory effect. Bioorganic chemistry 2019, 86, 501–506. Kim, C. S.; Kwon, O. W.; Kim, S. Y.; Kim, K. H.; Lee, K. R., A New Cyclic Triterpene Saponin from Phyteuma Japonicum. Heterocycles 2014, 89, 1913–1922. Kim, C. S.; Subedi, L.; Oh, J.; Kim, S. Y.; Choi, S. U.; Lee, K. R., A New Phenolic Compound from Salix Glandulosa. Heterocycles 2018, 96, 935–946. Yu, J. S.; Baek, J.; Park, H. B.; Moon, E.; Kim, S. Y.; Choi, S. U.; Kim, K. H., A new rearranged eudesmane sesquiterpene and bioactive sesquiterpenes from the twigs of Lindera glauca (Sieb. et Zucc.) Blume. Archives of Pharmacal Research 2016, 39, 1628–1634. Kim, K. H.; Lee, I. K.; Choi, S. U.; Lee, J. H.; Moon, E.; Kim, S. Y.; Lee, K. R., New triterpenoids from the tubers of Corydalis ternata: structural elucidation and bioactivity evaluation. Planta Medica 2011, 77, 1555–1558. Kim, K. H.; Moon, E.; Kim, H. K.; Oh, J. Y.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Phenolic constituents from the rhizomes of Acorus gramineus and their biological evaluation on antitumor and anti-inflammatory activities. Bioorg Med Chem Lett 2012, 22, 6155–6159. Park, K. J.; Cha, J. M.; Subedi, L.; Kim, S. Y.; Lee, K. R., Phenolic constituents from the twigs of Betula schmidtii collected in Goesan, Korea. Phytochemistry 2019, 167, 112085. Woo, K. W.; Suh, W. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Kim, K. H.; Lee, K. R., Phenolic Derivatives from the Stems of Lagerstroemia Indica and Their Biological Activity. Heterocycles 2015, 91, 2355–2366. Kim, C. S.; Kwon, O. W.; Kim, S. Y.; Choi, S. U.; Kim, J. Y.; Han, J. Y.; Choi, S. I.; Choi, J. G.; Kim, K. H.; Lee, K. R., Phenolic Glycosides from the Twigs of Salix glandulosa. Journal of Natural Products 2014, 77, 1955–1961. Cha, J. M.; Kim, D. H.; Lee, T. H.; Subedi, L.; Kim, S. Y.; Lee, K. R., Phytochemical constituents of Capsella bursa-pastoris and their anti-inflammatory activity. Natural Product Sciences 2018, 24, 132–138. Lee, S. Y.; Moon, E.; Kim, S. Y.; Lee, K. R., Quinic acid derivatives from Pimpinella brachycarpa exert anti-neuroinflammatory activity in lipopolysaccharide-induced microglia. Bioorganic Medicinal Chemistry Letters 2013, 23, 2140–2144. Kim, C. S.; Oh, J.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Rare Thioglycosides from the Roots of Wasabia japonica. Journal of Natural Products 2018, 81, 2129–2133. Suh, W. S.; Kwon, O. K.; Lee, T. H.; Subedi, L.; Kim, S. Y.; Lee, K. R., Secoiridoid Glycosides from the Twigs of Ligustrum obtusifolium Possess Anti-inflammatory and Neuroprotective Effects. Chemical Pharmaceutical Bulletin 2018, 66, 78–83. Park, K. J.; Kim, C. S.; Khan, Z.; Oh, J.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Securinega Alkaloids from the Twigs of Securinega suffruticosa and Their Biological Activities. Journal of Natural Products 2019, 82, 1345–1353. Van Kiem, P.; Hang, D. T.; Nhiem, N. X.; Tai, B. H.; Anh, H. L. T.; Van Cuong, P.; Quang, T. H.; Van Minh, C.; Van Dau, N.; Kim, Y.-A.; Subedi, L.; Kim, S. Y.; Kim, S. H., Sesquiterpene derivatives from marine sponge Smenospongia cerebriformis and their anti-inflammatory activity. Bioorganic medicinal chemistry letters 2017, 27, 1525–1529. Kim, C. S.; Kim, S. Y.; Moon, E.; Lee, M. K.; Lee, K. R., Steroidal constituents from the leaves of Hosta longipes and their inhibitory effects on nitric oxide production. Bioorganic Medicinal Chemistry Letters 2013, 23, 1771–1775. Kim, C. S.; Oh, J.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Structural Characterization of Terpenoids from Abies holophylla Using Computational and Statistical Methods and Their Biological Activities. Journal of Natural Products 2018, 81, 1795–1802. Yuan, Y.; Subedi, L.; Lim, D.; Jung, J.-K.; Kim, S. Y.; Seo, S.-Y., Synthesis and anti-neuroinflammatory activity of N-heterocyclic analogs based on natural biphenyl-neolignan honokiol. Bioorganic Medicinal Chemistry Letters 2019, 29, 329–333. Shobha, D.; Chari, M. A.; Mukkanti, K.; Kim, S. Y., Synthesis and anti-neuroinflammatory activity studies of substituted 3,4-dihydroquinoxalin-2-amine derivatives. Tetrahedron Letters 2012, 53, 2675–2679. Jin, G. H.; Ha, S. K.; Park, H. M.; Kang, B.; Kim, S. Y.; Kim, H.-D.; Ryu, J.-H.; Jeon, R., Synthesis of azaisoflavones and their inhibitory activities of NO production in activated microglia. Bioorg Med Chem Lett 2008, 18, 4092–4094. Lee, T. H.; Khan, Z.; Kim, S. Y.; Lee, K. R., Thiohydantoin and Hydantoin Derivatives from the Roots of Armoracia rusticana and Their Neurotrophic and Anti-neuroinflammatory Activities. Journal of Natural Products 2019, 82, 3020–3024. Woo, K. W.; Park, J. E.; Cha, J. M.; Subedi, L.; Kim, S. Y.; Lee, K. R., Three New Lignan Glycosides from the Firmiana simplex. Chemical Pharmaceutical Bulletin 2019, 67, 18–22. Kim, C. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Choi, S. Z.; Son, M. W.; Kim, K. H.; Lee, K. R., Two new phenolic compounds from the white flower of Impatiens balsamina. Phytochemistry Letters 2015, 14, 215–220. Suh, W. S.; Lee, S. Y.; Park, J. E.; Kim, D. H.; Kim, S. Y.; Lee, K. R., Two New Steroidal Alkaloids from the Bulbs of Fritillaria Thunbergii. Heterocycles 2018, 96, 925–934. Tuan Anh, H. L.; Kim, D.-C.; Ko, W.; Ha, T. M.; Nhiem, N. X.; Yen, P. H.; Tai, B. H.; Truong, L. H.; Long, V. N.; Gioi, T. J. P. b., Anti-inflammatory coumarins from Paramignya trimera. 2017, 55, 1195–1201. Yoon, C.-S.; Kim, K.-W.; Lee, S.-C.; Kim, Y.-C.; Oh, H. J. B.; letters, m. c., Anti-neuroinflammatory effects of sesquiterpenoids isolated from Nardostachys jatamansi. 2018, 28, 140–144. Li, N.; Wang, Y.; Li, X.; Zhang, H.; Zhou, D.; Wang, W.; Li, W.; Zhang, X.; Li, X.; Hou, Y. J. B.; Letters, M. C., Bioactive phenols as potential neuroinflammation inhibitors from the leaves of Xanthoceras sorbifolia Bunge. 2016, 26, 5018–5023. Chen, G.; Xie, Y.; Zhou, D.; Yang, Y.; Liu, J.; Hou, Y.; Cheng, M.; Liu, Y.; Li, N. J. P., Chemical constituents from shells of Xanthoceras sorbifolium. 2020, 172, 112288. Lee, S. R.; Kim, M. S.; Kim, S.; Hwang, K. W.; Park, S. Y. J. C.; Biodiversity, Constituents from Scutellaria barbata inhibiting nitric oxide production in LPS-stimulated microglial cells. 2017, 14, e1700231. Kim, D.-C.; Quang, T. H.; Ngan, N. T. T.; Yoon, C.-S.; Sohn, J. H.; Yim, J. H.; Feng, Y.; Che, Y.; Kim, Y.-C.; Oh, H. J. J. o. n. p., Dihydroisocoumarin derivatives from marine-derived fungal isolates and their anti-inflammatory effects in lipopolysaccharide-induced BV2 microglia. 2015, 78, 2948–2955. Ock, J.; Han, H. S.; Hong, S. H.; Lee, S. Y.; Han, Y. M.; Kwon, B. M.; Suk, K. J. B. j. o. p., Obovatol attenuates microglia-mediated neuroinflammation by modulating redox regulation. 2010, 159, 1646–1662. Yeh, C. H.; Yang, M. L.; Lee, C. Y.; Yang, C. P.; Li, Y. C.; Chen, C. J.; Kuan, Y. H. J. E. t., Wogonin attenuates endotoxin-induced prostaglandin E2 and nitric oxide production via Src‐ERK1/2‐NFκB pathway in BV‐2 microglial cells. 2014, 29, 1162–1170. Kim, K.-W.; Quang, T. H.; Ko, W.; Kim, D.-C.; Yoon, C.-S.; Oh, H.; Kim, Y.-C. J. P. b., Anti-neuroinflammatory effects of cudraflavanone A isolated from the chloroform fraction of Cudrania tricuspidata root bark. 2018, 56, 192–200. Schrödinger Maestro, Schrödinger, LLC, New York, 2019. Axen, S. D.; Huang, X.-P.; Cáceres, E. L.; Gendelev, L.; Roth, B. L.; Keiser, M. J., A Simple Representation of Three-Dimensional Molecular Structure. Journal of Medicinal Chemistry 2017, 60, 7393–7409. Duan, J.; Dixon, S. L.; Lowrie, J. F.; Sherman, W., Analysis and comparison of 2D fingerprints: insights into database screening performance using eight fingerprint methods. J Mol Graph Model 2010, 29, 157–170. Sastry, M.; Lowrie, J. F.; Dixon, S. L.; Sherman, W., Large-scale systematic analysis of 2D fingerprint methods and parameters to improve virtual screening enrichments. Journal of Chemical Information Modeling 2010, 50, 771–784. Schrödinger Canvas, Schrödinger, LLC, New York, NY, 2019. Schrödinger LigPrep, Schrödinger, LLC, New York, 2019. Schrödinger ConfGen, Schrödinger, LLC, New York, 2019. Watts, K. S.; Dalal, P.; Murphy, R. B.; Sherman, W.; Friesner, R. A.; Shelley, J. C., ConfGen: a conformational search method for efficient generation of bioactive conformers. Journal of Chemical Information Modeling 2010, 50, 534–546. Hawkins, P. C.; Skillman, A. G.; Warren, G. L.; Ellingson, B. A.; Stahl, M. T., Conformer generation with OMEGA: algorithm and validation using high quality structures from the Protein Databank and Cambridge Structural Database. Journal of Chemical Information Modeling 2010, 50, 572–584. OpenEye OMEGA 3.0.1.2 , OpenEye Scientific Software, Santa Fe, NM http://www.eyesopen.com ., 2018. Hemmerich, J.; Asilar, E.; Ecker, G. F., COVER: conformational oversampling as data augmentation for molecules. Journal of Cheminformatics 2020, 12, 1–12. Lama, P. D.; Kumar, S.; Kim, K.; Ahn, S.; Kim, M. H., Quantitative Prediction on the Enantioselectivity of Multiple Chiral Iodoarene Scaffolds Based on Whole Geometry. 2021, arXiv preprint arXiv:2103.14065. Abadi, M.; Barham, P.; Chen, J.; Chen, Z.; Davis, A.; Dean, J.; Devin, M.; Ghemawat, S.; Irving, G.; Isard, M. Tensorflow: A system for large-scale machine learning. In 12th {USENIX} symposium on operating systems design and implementation ({OSDI} 16), 2016; 2016; pp 265–283. Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V. J. t. J. o. m. L. r., Scikit-learn: Machine learning in Python. 2011, 12, 2825–2830. Taqi, A. M.; Awad, A.; Al-Azzo, F.; Milanova, M., The impact of multi-optimizers and data augmentation on TensorFlow convolutional neural network performance. In 2018 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR) 2018, 140. Algamal, Z. Y.; Lee, M. H., Penalized logistic regression with the adaptive LASSO for gene selection in high-dimensional cancer classification. Expert Systems with Applications 2015, 42, 9326–9332. Feng, Z. Z.; Yang, X.; Subedi, S.; McNicholas, P. D., The LASSO and sparse least squares regression methods for SNP selection in predicting quantitative traits. IEEE/ACM transactions on computational biology 2011, 9, 629–636. Zou, H.; Hastie, T., Regularization and variable selection via the elastic net. Journal of the royal statistical society: series B 2005, 67, 301–320. Xu, S. Bayesian Naïve Bayes classifiers to text classification. Journal of Information Science, 2018, 44(1), 48–59. Chicco, D.; Jurman, G., The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation. BMC Genomics 2020, 21, 6. Matthews, B. W. J. B. e. B. A.-P. S., Comparison of the predicted and observed secondary structure of T4 phage lysozyme. 1975, 405, 442–451. Powers, D. M., Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation. arXiv preprint 2020. Araujo, M. B., Pearson, R. G., Thuiller, W., & Erhard, M. (2005). Validation of species–climate impact models under climate change. Global change biology, 11(9), 1504–1513. Lee, S. H.; Ahn, S.; Kim, M. H., Comparing a query compound with drug target classes using 3D-chemical similarity. International Journal of Molecular Sciences, 2020, 21(12), 4208. Ahn, S.; Lee, S. E.; Kim, M. H.. Random-forest model for drug–target interaction prediction via Kullback–Leibler divergence. Journal of Cheminformatics, 2022, 14(1), 1–13. Aizawa, A., An information-theoretic perspective of tf–idf measures. Information Processing & Management, 2003, 39(1), 45–65. Additional Declarations No competing interests reported. Supplementary Files SupplementarytablesSciRep.docx Cite Share Download PDF Status: Published Journal Publication published 16 Nov, 2024 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 02 Jul, 2024 Reviews received at journal 30 Jun, 2024 Reviews received at journal 24 Jun, 2024 Reviewers agreed at journal 22 Jun, 2024 Reviewers agreed at journal 21 Jun, 2024 Reviews received at journal 08 Apr, 2024 Reviewers agreed at journal 28 Mar, 2024 Reviewers invited by journal 28 Mar, 2024 Editor assigned by journal 13 Feb, 2024 Editor invited by journal 28 Dec, 2023 Submission checks completed at journal 28 Dec, 2023 First submitted to journal 27 Dec, 2023 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3812369","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":264546371,"identity":"afc9752a-f984-46b0-bd44-b860908399ad","order_by":0,"name":"Si Eun Lee","email":"","orcid":"","institution":"Gachon University","correspondingAuthor":false,"prefix":"","firstName":"Si","middleName":"Eun","lastName":"Lee","suffix":""},{"id":264546372,"identity":"e1d48a1d-3cec-4a36-9a39-7342c8b514ae","order_by":1,"name":"Sangjin Ahn","email":"","orcid":"","institution":"Ajou University","correspondingAuthor":false,"prefix":"","firstName":"Sangjin","middleName":"","lastName":"Ahn","suffix":""},{"id":264546373,"identity":"0c9d73e3-7aef-425c-8bd5-f707cc083810","order_by":2,"name":"Surendra Kumar","email":"","orcid":"","institution":"Gachon University","correspondingAuthor":false,"prefix":"","firstName":"Surendra","middleName":"","lastName":"Kumar","suffix":""},{"id":264546374,"identity":"8e7084b0-3e1c-4180-b622-6e19239f1eea","order_by":3,"name":"Mi-hyun Kim","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA70lEQVRIiWNgGAWjYBACCQYehsM/DP7JAdlsUDE2fBrAWhgfMxQcMCZJC7Mxw4cDiQ1Ea5GcdvaYdIHBnfT57oefPfi4g0Gev4Et7QM+LdLSeWnSMwye5W48k2ZuOPMMg+GMA2yHZ+DTIiedYybBY8Ccu7Ehh02at42BcQMDezNeh8G0pBv2vwFrsSeoRVo6x9iYx+BwgrwExJbEDQxsh/FqkZydY/hwhkGa4QaJZ2aSM9skkmccZkvGq0Xido7BgQ9/bOTl+5OfSXxss7Htb28zxqsFDgwOQIxgYGAmTgMDg3wDsSpHwSgYBaNgxAEAKzdDEXGa0L4AAAAASUVORK5CYII=","orcid":"","institution":"Gachon University","correspondingAuthor":true,"prefix":"","firstName":"Mi-hyun","middleName":"","lastName":"Kim","suffix":""}],"badges":[],"createdAt":"2023-12-27 13:14:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3812369/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3812369/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-024-78823-3","type":"published","date":"2024-11-16T15:58:15+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":49019717,"identity":"711ac1b3-c533-44f7-b267-b8e37dcd198e","added_by":"auto","created_at":"2024-01-01 08:09:06","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":177209,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eAnti-neuroinflammatory potency\u003c/em\u003e distribution of dataset\u003c/p\u003e","description":"","filename":"Fig1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3812369/v1/d569972f91024412673b1ca2.jpg"},{"id":49019715,"identity":"0a9dd529-a88d-46d8-8834-3491e8ac0b70","added_by":"auto","created_at":"2024-01-01 08:09:06","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":106597,"visible":true,"origin":"","legend":"\u003cp\u003eLearning architectures of neural networks used in this study: (a) MLP, (b) RNN, and (c) CNN\u003c/p\u003e","description":"","filename":"Fig2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3812369/v1/2409615809e8c027ac3d43f7.jpg"},{"id":49019719,"identity":"900991dc-a898-497a-aa75-27379a07cc95","added_by":"auto","created_at":"2024-01-01 08:09:07","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":115523,"visible":true,"origin":"","legend":"\u003cp\u003eScaffold analysis of dataset\u003c/p\u003e","description":"","filename":"Fig3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3812369/v1/349c8c01f42c9487db25addc.jpg"},{"id":49019716,"identity":"108c6ad9-08cf-438e-ac35-d5ca256ace08","added_by":"auto","created_at":"2024-01-01 08:09:06","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":63525,"visible":true,"origin":"","legend":"\u003cp\u003e3D-Molecular representation and featurization process to generate location information\u003c/p\u003e","description":"","filename":"Fig4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3812369/v1/e8bcf447e45228f47a1b5092.jpg"},{"id":49019721,"identity":"0e38cfd1-c7cb-4326-a672-88eb76a1ffc1","added_by":"auto","created_at":"2024-01-01 08:09:07","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":55019,"visible":true,"origin":"","legend":"\u003cp\u003eThe AUC distribution of out-of-set compounds in Neural Network\u003c/p\u003e","description":"","filename":"Fig5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3812369/v1/edabadeaa87c0ebba8e578c9.jpg"},{"id":49019718,"identity":"250ed073-8d83-4c5c-8de8-daaa4206c8e7","added_by":"auto","created_at":"2024-01-01 08:09:07","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":202122,"visible":true,"origin":"","legend":"\u003cp\u003eDiscrimination between active and inactive of third dataset by the classification models. (a) Coumarin, (b) Flavonoid, (c) Saponin\u003c/p\u003e","description":"","filename":"Fig6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-3812369/v1/c83cf8a08282ecf6b718b609.jpg"},{"id":69285329,"identity":"e26e9993-db29-4d81-a4da-f11be141dac9","added_by":"auto","created_at":"2024-11-18 19:25:13","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1267348,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3812369/v1/405550b1-348d-4d41-800e-8516032ebd38.pdf"},{"id":49019720,"identity":"4cabb756-c23b-468b-b487-8a577e75ec7d","added_by":"auto","created_at":"2024-01-01 08:09:07","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":242553,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementarytablesSciRep.docx","url":"https://assets-eu.researchsquare.com/files/rs-3812369/v1/46eb54c5d680ebb4700b11c2.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"NO-Classifier: Prediction of Anti- Neuroinflammatory Agents Using Text Mining of 3D Molecular Fingerprints","fulltext":[{"header":"Highlights","content":"\u003cul class=\"decimal_type\"\u003e\n \u003cli\u003eNO-Classifier is a phenotype-structure relationship model \u0026nbsp;\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eNO-Classifier predicts an anti-neuroinflammatory potency of natural and non-natural compounds\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eNO inhibitory potency in hyperactivated microglia was trained with\u0026nbsp;the location information of 3D molecular structures\u003c/li\u003e\n \u003cli\u003eThe statistical performance of NO-Classifier\u0026nbsp;with\u0026nbsp;\u003cem\u003eaccuracy of 0.969 and AUC of 0.996\u003c/em\u003e\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"1. Introduction","content":"\u003cp\u003eThe lack of clarity in basic biological and pathological mechanisms related to brain has made CNS drug discovery difficult. Despite the difficulty, some CNS drugs have been developed based on phenotypic effects. Neuroinflammation is one of hallmark phenotypes observed in neurological disorders1, which include age-related dementia2, neurodegenerative diseases such as amyotrophic lateral sclerosis (ALS), Parkinson's disease (PD), and Alzheimer's disease (AD)3\u0026ndash;5. Neuroinflammation has been well studied in cellular level. For example, lipopolysaccharide (LPS) can induce the inflammation of neuronal cells through the hyperactivation of microglia which plays proinflammatory roles in CNS6. Moreover, because LPS does not involve CNS disruption, it is used as a stimulant to generate neuroinflammation. The hyperactivated microglia generated inflammatory substances such as nitric oxide (NO). The NO level in cells contributes to the balance between pro-inflammation and anti-inflammation7. Thus, NO production in LPS induced microglial cells can be used for measuring neuroinflammation of cells and NO production assay is one of the simple screening methods for anti-neuroinflammatory agents.\u003c/p\u003e \u003cp\u003eBecause NO inhibitory potency of a testing compound observed in the assay is an effective indicator to show how much the testing compound can suppress neuroinflammation, we judged that the testing data of NO production assay can be used for a phenotype-structure relationship model. Surely, the model should be built using phenotype-active or inactive compounds without any information on specific targets. When the phenotype-structure relationship model is realized, the chemo-centric study can aid the drug design and drug screening for neuroinflammation-related diseases such as ALS, PD, and AD. This motivation made us developed a classification model to predict anti-inflammatory potency of various compounds of \u0026gt;\u0026thinsp;29 multiple scaffolds using their valid IC50 values of NO assay. Our sequential research studies on 3D chemical similarity and diverse molecular featurization for drug discovery encouraged to investigate molecular features for this phenotype model. In this study, we report the location information of 3D molecular structure as an efficient molecular feature and the 3D location information based classification model using neural network and conventional classifiers.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Dataset Selection and Preparation\u003c/h2\u003e \u003cp\u003eWe collected academic papers including compounds related with neuroinflammation using Google Scholar and EMBASE. The criteria of the collection were structural diversity of the compounds and the balance between natural occurring and artificial synthesizing. Moreover, the artificial synthetic molecules were mainly in-house data accumulated in our serial studies for neuroinflammation \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e8,9\u003c/span\u003e. While 95 papers of 122 papers allowed us to extract compounds structures and their biological activities through Reaxys and Pubchem. The collected compounds were sorted by their pharmacological effects: 584 compounds for inhibiting NO production, 197 compounds for inducing nerve growth factor, 5 compounds for promoting neurite outgrowth, and fewer compounds for other functions such as ameliorating neurotoxicity. Surely, the number of NO-inhibiting compounds was dominant, and thus we chose NO assay to describe drugs\u0026rsquo; effect on neuroinflammation and used the corresponding data set (\u003cb\u003eSupplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e-S2\u003c/b\u003e). When a compound has multiple IC50 values, the largest IC50 value was chosen among them. After removing 30 compounds owing to their severe toxicity, we gained the IC50 distribution of the dataset and decided the threshold (IC50 cut-off: 37 \u0026micro;M) values between active and inactive in the distribution (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). In summary, 584 compounds of 65 papers were used for model training and validation 10\u0026ndash;74. Moreover, another dataset was collected and prepared to use out-of-set 75\u0026ndash;83.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 3D-Molecular Representation and Featurization\u003c/h2\u003e \u003cp\u003eIn order to generate 3D molecular representation, 3D conformers should be priori generated and sampled using Schr\u0026ouml;dinger Suite 84. We intended to calculate 3D molecular fingerprints85 from 3D conformers. Our data matrix comprising the SMILES of the collected compounds, their respective names, IC50 values, and other activities was manipulated using MAESTRO and Canvas of Schr\u0026ouml;dinger Suite86-88 and each SMILES was transformed to 3D conformation using LigPrep89. Multiple conformers were further generated using ConfGen90\u003csup\u003e,\u003c/sup\u003e 91. Different numbers of rotatable bonds present in various compounds still made the multiple conformers cannot enough represent the conformational space of flexible compounds. Thus, conformer ensembles were more sampled based on the generated multiple conformers using Omega of Openeye 92\u003csup\u003e,\u003c/sup\u003e 93 employing the default parameters. Finally, 60,275 3D conformers of 544 compounds were successfully acquired. The conformer ensemble of 60,275 size was employed to enhance diversity and scale of the dataset based on the literature 94, where conformational oversampling enhanced a dataset by generation of multiple conformations of a molecule and gained a desirable balance between sensitivity and specificity.\u003c/p\u003e \u003cp\u003eSubsequently, E3FP of 1024-bit length as a 3D molecular fingerprint was generated from the conformer ensemble. E3FP encodes three-dimensional molecular geometry regardless of chemical bonds\u0026rsquo; presence or absence (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/keiserlab/e3fp)8\u003c/span\u003e\u003cspan address=\"https://github.com/keiserlab/e3fp)8\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e5. Upon computation of the 3D fingerprint, extant substructures are encoded to one (on-bit) within the fixed 1024-bits and some absence is transformed to zero (off-bit) in the array (as depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), thereby obtaining relative locational information of atoms95. While E3FP can be generally used as a sparse bitvector or bitstring, herein we tried to modify the sparse 1024 bit into the compact array of irregular length (the maximum length: 212). In detail, the location of every on-bit information was written to our positional array. And then we tokenized every element in the positional array using Keras Tokenizer96 and vectorized the relative location information of on-bit into an integer list using TF-IDF Vectorizer from scikit-learn 97. The above process applied to the third set as well as training/test set. Finally, we used the vectorized \"positional information\" as feature set for the two type bi-class classification models, which are neural networks using Keras and conventional machine learning using scikit-learn, and classification analysis to choose the best NO classifier model in Jupyter Notebook environment.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Model training\u003c/h2\u003e \u003cp\u003eAs the above mentioned, Keras/TensorFlow and scikit-learn were chosen among the ideal Python libraries for machine learning96\u003csup\u003e,\u003c/sup\u003e 97 in order to train NO inhibitory function of drugs and their 3D structures with reliability and efficacy. Based on three neural networks (NN) architectures and four classifier analysis (CA) methods, a total of seven discrete models were built. The construction and training of all models were executed in the condition of randomly chosen 80% of the dataset, the external validation was performed with the remaining 20% as test sets.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e2.3.1 Neural Network\u003c/h2\u003e \u003cp\u003eWe built three \u003cem\u003eNN\u003c/em\u003e architectures of MLP, convolutional neural network (CNN), and recurrent neural network (RNN) to predict NO inhibitory potency of testing drugs (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The initial parameters for the \u003cem\u003eNN\u003c/em\u003e architectures were tested with the epoch number of 15, batch size of 256, and learning rate of 0.001. The MLP and RNN have one dense layer with 32 units for each model, whereas the CNN has two dense layers. In detail, the MLP layer parameters (1024, 32, 212) define the vocabulary size, the dimensionality of the embedding space, and the maximum sequence length of the input text data, respectively. In other words, the maximum word index of the location information was 1024, which was the vocabulary size for word indexing from 0 to 1023. Each word was converted into a 32-dimensional vector. The maximal sequence length of the input drugs was 212 words.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eSubsequently, the flatten layer transforms the multidimensional input of shape (None, 212, 32) into a one-dimensional vector of shape (None, 6784), effectively collapsing all spatial dimensions. Lastly, the dense layer outputs a single value per batch. Similarly, after the same MLP, the RNN model incorporated a long short-term memory (LSTM) layer, which is designed for retaining information over extended sequences, effectively replacing the flatten layer. In addition to the aforementioned architectures, the CNN model incorporated a GlobalMaxPool1D layer for pooling operations, followed by two dense layers employing ReLU and sigmoid activations, respectively. All three NN models used binary cross-entropy as the loss function to calculate the gradients, and the RMSprop algorithm, an adaptive variant of the standard gradient descent, to optimize the parameters 98. Owing to the intrinsic stochasticity of neural network training, each model was preserved in the H5 file format. Under the condition, the variations in AUC values were contained within a narrow margin, not exceeding 0.1.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e2.3.2 Classifier Analysis\u003c/h2\u003e \u003cp\u003eIn this study, we tested the performance of several CA techniques, including logistic classification, ridge classification, least absolute shrinkage selection operator (LASSO), and Na\u0026iuml;ve Bayes. Evert model employed the TF-IDF Vectorizer to transform the location information into a matrix of inverse frequency features for each word. The logistic classifier was initialized with a random state of 0 as the seed value for the data shuffling process. Ridge Classifier was chosen based on the characteristics of the method, L2 penalty to address multicollinearity and feature selection to identify the most relevant variables for the classification task. Similarly, the LASSO classifier performed feature subset based on the convex optimization (L1 penalty)100\u003csup\u003e,\u003c/sup\u003e 101. Na\u0026iuml;ve Bayes classifier is known to be an effective machine-learning method for text mining. The rapid and advanced method has been applied to various real-world scenarios, including spam filtering, document categorization, and text classification. Na\u0026iuml;ve Bayes classifier offers a range of models, including the Multinomial, Gaussian, and Bernoulli models, which Bernoulli and Gaussian event models implemented in scikit learn are not fully Bayesian 102. We developed a Gaussian Na\u0026iuml;ve Bayes (Gaussian NB) classifier to extract features, including word counts.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Model validation\u003c/h2\u003e \u003cp\u003eTo develop the optimal predictive model, the built models was evaluated by several statistical metrics of Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e such as mean accuracy, specificity, sensitivity, F1-Score, area under the curve (AUC), and Matthews Correlation Coefficient (MCC), which is a particular case of the phi coefficient, ϕ, with a value between \u0026minus;\u0026thinsp;1 and 1103\u003csup\u003e,\u003c/sup\u003e 104. Moreover, the confusion matrices of models were directly compared for the performance evaluation. The common statistical metrics for evaluating classification models were mathematically defined in the Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. In this study, ACC, F1-score, AUC, MCC were mainly used to compare the performance between models. Obviously, ACC is conventional metric but cannot properly explain models built from imbalanced dataset between active and inactive. To compensate the limitation, F1-score and MCC were chosen and MCC was preferred for us. In general, MCC values more than 0.7 indicate the significant correlation between the predicted and actual classification. MCC values in the range of 0.4 to 0.7 for the MCC suggest a moderate level of the correlation, ranging from 0.4 to 0.7. Theoretically, while MCC of zero would result in a random prediction, MCC values ranging from 0 to 0.4 still have the correlation of low strength but superior to the random and MCC values below 0 indicate the inverse correlation104. (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Clearly, MCC is a contingency matrix technique used to determine the Pearson product-moment correlation coefficient. Therefore, it has the same interpretation as Pearson's r correlation105. Moreover, AUC was calculated from receiver operator characteristic (ROC) curves, which is plot of the true positive rate against the false positive rate. In detail, we used AUC values to judge discriminative potential of models: excellent (\u0026gt;\u0026thinsp;0.90), good (\u0026gt;\u0026thinsp;0.80), fair (\u0026gt;\u0026thinsp;0.70), and poor discrimination (\u0026lt;\u0026thinsp;0.70)106.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eEvaluation metrics for the classification models.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetrics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEquation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSensitivity (SE)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\frac{TP}{TP+FN}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpecificity (SP)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\frac{TN}{FP+TN}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccuracy (ACC)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\frac{TP+TN}{TP+TN+FP+FN}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF1 score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\frac{2TP}{2TP+FP+FN}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMCC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\frac{TP*TN-FP*FN}{\\sqrt{\\left(\\left(TP+FP\\right)*\\left(TP+FN\\right)*\\left(TN+FP\\right)*\\left(TN+FN\\right)\\right)}}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eN: total number of dataset, TP: true positives, TN: true negatives, FN: false negatives, and FP: false positives.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results and Discussion","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Characterization of Anti-Inflammatory Dataset\u003c/h2\u003e \u003cp\u003eWe collected anti-inflammatory agents and their activity information based on cell-based assay from academic articles10-74, Reaxys, EMBASE, and Pubchem. After the data regularization, we found NO production assay, which measure the intensity of pro-inflammation in hyperactivated microglia, has the massive data in the terms of scaffold diversity (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Therefore, the assay data as dataset was selected, and the IC50 value of NO production assay was used to describe anti-inflammatory potency of testing drugs. Before building model, the characterization of the dataset was performed through the scaffold analysis of compounds in the dataset and distribution of IC50 values. Under the IC50 cut-off of 37.1, the ratio between active and inactive was 1: 1. Moreover, 27 compounds of the same assay (LPS-induced NO assay) were further collected and used as the out-of-set to validate the models 75\u0026ndash;83. The third set compounds were separated to csv files according to the scaffolds such as steroid, lignan, phenylpropanoid, iridoid, coumarin, and flavonoid.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Text Mining Based Molecular Featurization\u003c/h2\u003e \u003cp\u003eHerein, we intended to predict the molecular modulation of neuroinflammation by the three \u003cem\u003eNN\u003c/em\u003e and four \u003cem\u003eCA\u003c/em\u003e models. We named the models NO-Classifier. In detail, the primary aim of the built models is to explain the correlation of the 3D structures of testing compounds with their NO inhibitory potency and to predict potency of new compounds using their 3D structures. The final aim is to design new competent modulators for neuroinflammation treatment using the predictive models. For the purpose, E3FP location information was chosen for molecular featurization of the 3D structures (exactly, 3D molecular geometry). In detail, 3D Molecular representation (E3FP of 1024-bit length) was generated from conformer ensemble (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;60,275 of 544 compounds), which was prepared through conformational oversampling to enhance diversity and scale of the dataset based on the literature 94. It is well-known that one 2D molecular fingerprint can be decoded to one unique 2D molecule but 3D molecular fingerprint cannot be perfectly decoded to one 3D conformation85. Moreover, how to use enormous 3D conformational space has been one huddle in drug design and screening \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e107,108\u003c/span\u003e.Thus, although E3FP, itself has been directly used as molecular descriptor, we intended to compress the information in E3FP for the cost effectiveness of the 3D data. At least, this trial reduced the absolute size of the 3D data. The transformation of sparse E3FP bitvector was conducted through gathering positional information of only present substructures (on-bit). As the result, 3D molecular geometry of each conformer was written to the dense array (of Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) like text data of irregular length. The relative location information of atom level substructure was applied to natural language processing (NLP) tools. In other words, we treated each value in the dense array as each word in a sentence. Because the conformers of 60,275 presented the maximum length of 212, the maximum length is in the range of general sentences for NLP. Thus, we tested some NLP tools such as NLTK (Natural Language Toolkit), Keras Processing, Word2vec, and Doc2vec. Conventional NLP follows the process consisting of tokenization, vectorization, and so on for feature MLP. We also tokenized every element of the positional array using Keras Tokenizer 96 and vectorized the relative location information of on-bit into the transformed inverse frequency feature using TF-IDF Vectorizer as a probability weighted information, 97,109. Finally, we used the vectorized \"positional information\" as feature set for our classification models.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Performance Evaluation of NO-Classifier\u003c/h2\u003e \u003cp\u003e \u003cem\u003eNN\u003c/em\u003e and \u003cem\u003eCA\u003c/em\u003e models were built from the NLP treated feature of positional information and their predictive power for the molecular modulation of neuroinflammation was evaluated by several statistical metrics as shown in the method section (2.4 Model Validation). The Tables\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e describes the internal 10-fold validation results using training data (80% of the dataset) and external validation using test data (20% of of the dataset) under the random dividing condition with specific seed numbers. Every NNs showed superior performance to CA models in the validation. Particularly, CNN demonstrated the best performance in terms of every statistical metric (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePrediction performance of classification models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"13\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c13\" colnum=\"13\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eML\u003c/p\u003e \u003cp\u003eMethod\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"6\" nameend=\"c7\" namest=\"c2\"\u003e \u003cp\u003eExternal Validation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"6\" nameend=\"c13\" namest=\"c8\"\u003e \u003cp\u003e10-Fold Internal Cross-Validation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eACC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1\u003c/p\u003e \u003cp\u003eScore\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eMCC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eACC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eSE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eF1\u003c/p\u003e \u003cp\u003eScore\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c12\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c13\"\u003e \u003cp\u003eMCC\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMLP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.962\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.968\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.952\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.952\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.994\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.920\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.983\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.984\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.981\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.979\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.999\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.965\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.966\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.973\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.955\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.957\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.994\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.928\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.982\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.985\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.976\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.977\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.999\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.962\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.969\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.973\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.961\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.961\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.996\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.935\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.983\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.984\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.982\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.980\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.999\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.966\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.886\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.899\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.865\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.855\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.951\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.761\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.892\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.902\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.876\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.865\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.958\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.775\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRidge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.884\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.897\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.864\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.853\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.948\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.758\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.887\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.899\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.869\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.859\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.954\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.764\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLASSO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.885\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.899\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.865\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.854\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.950\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.759\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.890\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.901\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.874\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.864\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.956\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.772\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNaive Bayes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.779\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.844\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.700\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.739\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.861\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.552\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.840\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.956\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.716\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.812\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.865\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.696\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"13\"\u003eACC: accuracy, SP: specificity, SE: sensitivity, F1-score: see equation in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, AUC: area under curve, MCC: Matthews\u0026rsquo;s correlation coefficient\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAmong logistic classifier, ridge classifier, LASSO classifier, and Na\u0026iuml;ve Bayes classifier, LASSO was the best CA model but it\u0026rsquo;s statistical metrics was slightly lower than those of \u003cem\u003eNN\u003c/em\u003e models. Notably, the AUC values of CA models satisfied the excellent criterion (\u0026gt;\u0026thinsp;0.90) with more than 0.95 except for Na\u0026iuml;ve Bayes classifier (AUC: 0.861), which still can be considered a commendable score106. Despite other statistical metrics inferior to \u003cem\u003eNN\u003c/em\u003e models, every CA model presented the significant correlation between the predicted and actual classification based on the MCC values of \u0026gt;\u0026thinsp;0.7 except for Na\u0026iuml;ve Bayes classifier (moderate significant correlation with MCC of 0.696 for 10-fold cross validation and 0.552 for external validation)104. In summary, the findings suggest that our NLP treated 3D molecular feature works properly to judge the anti-inflammatory potency within chemical space and conformational space of the data set. The results encouraged us to further investigate the proficiency of our NO-Classifier in identifying new molecular modulators for neuroinflammation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Out-of-Set Validation of NO-Classifier\u003c/h2\u003e \u003cp\u003eFurther investigation of NO-Classifier was designed using \u0026lsquo;out-of-set\u0026rsquo; data, which were not included in the dataset for either model training or validation. It is also called 3rd set. The new data can be grouped into two categories according to their sub-structures (scaffolds). If some new compound is absent in either training or test data but the scaffold of the compound is in the training or test data, the compound is included in the first category. Meanwhile, if both a new compound and its scaffold are absent in the training or test data, the compound is included in the second category. We expected the predictive performance of NO-Classifier to be retained for compounds in the first category. Meanwhile, we forecast that decreased performance for new scaffolds and wondered whether the second category compounds were compatible for NO-Classifier or not. As our expectation, the coumarin and flavonoid compounds (exact expression: derivatives) in the first category demonstrated promising outcomes concerning the AUC of the CNN (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e), with values of 0.913 (for coumarin) and 0.947 (for flavonoids). The compounds with saponin scaffold, which is absent in either the training data or test data, gave us the AUC value decreased to 0.760 in the CNN model. When we consider the dissimilarity of saponin derivatives with the dataset, the fair (\u0026gt;\u0026thinsp;0.70) AUC results of CNN model demonstrates the prediction potential of NO-Classifier applicable for future new coming compounds.\u003c/p\u003e \u003cp\u003eIn sequence, the out-of-set validation was performed for the CA models. In the case of flavonoid derivatives, the AUC values of logistic classifier (0.892) and LASSO classifier (0.867) exhibited good results (\u0026gt;\u0026thinsp;0.80) and Na\u0026iuml;ve Bayes classifier gave fair AUC of 0.718. Despite the first category of coumarins, CA models didn\u0026rsquo;t present fair AUC values. It suggests other factors except for common scaffolds. When flavonoid is compared with coumarin, the conformational rigidity could generate the limited number of conformers. Thus, we guess the rigidity of common scaffolds can make an effect on prediction performance. Our models tended to work more properly for compounds with a flexible scaffold able to have the high coverage for conformational space to predict future upcoming compounds. Remarkably, saponin, which was not incorporated into the training-test set and exhibited low similarity, demonstrated divergent outcomes in both CNN model and classical analysis models.\u003c/p\u003e \u003cp\u003eTo doubly confirm the performance of NO-Classifier, we produced decision regions (or decision boundary) for the CA models. The analysis with visualized maps revealed that the CA models exhibited a discerning ability towards flavonoid derivatives. Interestingly, despite the structural dissimilarity of saponin, decision regions also demonstrated effectively discerning performance. Once more, it is worth noting that almost compounds, which were gathered explicitly, fell into their right classification (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). Thus, the classification models and decision regions demonstrated reliable performance in effectively discerning the anti-inflammatory potency of compounds in inflammatory cell phenotype.\u003c/p\u003e \u003c/div\u003e"},{"header":"4. Conclusion","content":"\u003cp\u003eWe developed NO-Classifier using three NN and four CA algorithms, which is classification models for predicting anti-inflammatory potency against neuroinflammation. The models learned the 3D location information of multiple structures (conformer ensembles of drugs) as molecular feature and IC50 values of NO assay. NO-Classifier satisfied statistical metrics for both internal and external validation to exhibit the classification performance. Moreover, the satisfactory performance across all assessed metrics was doubly confirmed by both out-of-set validation and decision region analysis of the out-of-set. It demonstrates NO-Classifier can be used for forecasting anti-inflammatory potency of new coming compounds. At least, NO-Classifier can screen drugs to suppress neuroinflammatory process in neurodegenerative diseases. In the future, the prediction coverage of NO-Classifier can be expanded to more diverse and different compounds using the updated molecular data.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003eData Availability \u0026amp; Source Codes\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eSupplementary file is available.\u0026nbsp;All source code, and data are available in GitHub.\u003c/p\u003e\n\u003cp\u003ehttps://github.com/college-of-pharmacy-gachon-university/NO-Classifier\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThe webserver is available:\u0026nbsp;\u003c/em\u003ehttps://no-classifier.onrender.com\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eAcknowledgement \u0026amp; Funding\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe authors would like to express their gratitude to Professor Sun Yeou Kim and her laboratory for their invaluable suggestion and contribution to the initial data collection. The authors would like to thank OpenEye Scientific Software for providing an academic free license. This study was supported by the Basic Science Research Program of the National Research Foundation of Korea (NRF), which is funded by the Ministry of Education, Science and Technology (No.: 2022R1A2C2091810).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eAuthors\u0026rsquo; Contribution\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eM.K. conceived and designed the study. Under M. K.\u0026rsquo;s plan, J.A. and S.L. investigated diverse molecular featurization and S.L. fully conducted data collection and manipulation. M. K., and S. L. analyzed every data and result. S.K. refined source codes and developed webserver. M. K., and S. L. wrote the manuscript and revised it. M.K. provided the molecular modeling lab facility. All authors read and approved the final manuscript. The authors confirm that this article content has no conflicts of interest.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSchwartz, M.; Deczkowska, A., Neurological disease as a failure of brain\u0026ndash;immune crosstalk: the multiple faces of neuroinflammation. Trends in immunology 2016, 37, 668\u0026ndash;679.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLynch, M. A., Age-related neuroinflammatory changes negatively impact on neuronal function. Frontiers in Aging Neuroscience 2010, 1, 6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeneka, M. T.; O'Banion, M. K., Inflammatory processes in Alzheimer's disease. Journal of Neuroimmunology 2007, 184, 69\u0026ndash;91.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFrank-Cannon, T. C.; Alto, L. T.; McAlpine, F. E.; Tansey, M. G., Does neuroinflammation fan the flame in neurodegenerative diseases? Molecular neurodegeneration 2009, 4, 1\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeppner, F. L.; Ransohoff, R. M.; Becher, B., Immune attack: the role of inflammation in Alzheimer disease. Nature Reviews Neuroscience 2015, 16, 358\u0026ndash;372.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, Z.; Jalabi, W.; Shpargel, K. B.; Farabaugh, K. T.; Dutta, R.; Yin, X.; Kidd, G. J.; Bergmann, C. C.; Stohlman, S. A.; Trapp, B. D., Lipopolysaccharide-induced microglial activation and neuroprotection against experimental brain injury is independent of hematogenous TLR4. Journal of Neuroscience 2012, 32, 11706\u0026ndash;11715.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBourgognon, J.-M.; Spiers, J. G.; Robinson, S. W.; Scheiblich, H.; Glynn, P.; Ortori, C.; Bradley, S. J.; Tobin, A. B.; Steinert, J. R., Inhibition of neuroinflammatory nitric oxide signaling suppresses glycation and prevents neuronal dysfunction in mouse prion disease. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e 2021, 118.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVenkanna, A.; Kwon, O. W.; Afzal, S.; Jang, C.; Cho, K.; Yadav, D. K.; Kim, K.; Park, H.-g.; Chun, K.-H.; Kim, S. Y.; Kim, M.-h., Pharmacological use of a novel scaffold, anomeric N,N-diarylamino tetrahydropyran: molecular similarity search, chemocentric target profiling, and experimental evidence, Scientific Reports, 2017, 7, 12535\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVenkanna, A.; Subedi, L.; Teli, M. K.; Lama, P. D.; Nangunuri, B. G.; Lee, S. Y.; Kim, S.Y.; Kim, M. H., Positioning of an unprecedented spiro [5.5] undeca ring system into kinase inhibitor space. Scientific Reports, 2020, 10(1), 21265.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, D. H.; Kim, C. S.; Subedi, L.; Kim, S. Y.; Lee, K. R., Alkaloids of NIRAM, natural dye from Polygonum tinctorium, and their anti-inflammatory activities. Tetrahedron Letters 2019, 60, 151130.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Kim, C. S.; Park, Y. J.; Moon, E.; Choi, S. U.; Lee, J. H.; Kim, S. Y.; Lee, K. R., Anti-inflammatory and antitumor phenylpropanoid sucrosides from the seeds of Raphanus sativus. Bioorganic Medicinal Chemistry Letters 2015, 25, 96\u0026ndash;99.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSubedi, L.; Lee, J. H.; Yumnam, S.; Ji, E.; Kim, S. Y., Anti-inflammatory effect of sulforaphane on LPS-activated microglia potentially through JNK/AP-1/NF-κB inhibition and Nrf2/HO-1 activation. Cells 2019, 8, 194.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark, K. J.; Subedi, L.; Kim, S. Y.; Lee, K. R., Anti-inflammatory terpenoid derivatives from the twigs of Syringa oblata var. dilatata. Phytochemistry Letters 2018, 27, 183\u0026ndash;186.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Bae, M.; Oh, J.; Subedi, L.; Suh, W. S.; Choi, S. Z.; Son, M. W.; Kim, S. Y.; Choi, S. U.; Oh, D.-C., Anti-Neurodegenerative Biflavonoid Glycosides from Impatiens balsamina. Journal of Natural Products 2017, 80, 471\u0026ndash;478.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHa, S. K.; Shobha, D.; Moon, E.; Chari, M. A.; Mukkanti, K.; Kim, S.-H.; Ahn, K.-H.; Kim, S. Y., Anti-neuroinflammatory activity of 1,5-benzodiazepine derivatives. Bioorganic Medicinal Chemistry Letters 2010, 20, 3969\u0026ndash;3971.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, S. Y.; Suh, W. S.; Cha, J. M.; Moon, E.; Ha, S. K.; Kim, S. Y.; Lee, K. R., Anti-neuroinflammatory constituents from Sinomenium acutum rhizomes. Phytochemistry Letters 2016, 17, 79\u0026ndash;84.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Lee, K. H.; Kim, H. K.; Moon, E.; Kim, S.-H.; Kim, S. Y.; Kim, K. R.; Lee, K. R., Antineuroinflammatory constituents from the root extract of Paris verticillata. Canadian Journal of Chemistry 2011, 89, 441\u0026ndash;445.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWoo, K. W.; Moon, E.; Kwon, O. W.; Lee, S. O.; Kim, S. Y.; Choi, S. Z.; Son, M. W.; Lee, K. R., Anti-neuroinflammatory diarylheptanoids from the rhizomes of Dioscorea nipponica. Bioorganic Medicinal Chemistry Letters 2013, 23, 3806\u0026ndash;3809.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuh, W. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Bioactive lignan constituents from the twigs of Sambucus williamsii. Bioorganic Medicinal Chemistry Letters 2016, 26, 1877\u0026ndash;1880.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWoo, K. W.; Suh, W. S.; Subedi, L.; Kim, S. Y.; Kim, A.; Lee, K. R., Bioactive lignan derivatives from the stems of Firmiana simplex. Bioorganic Medicinal Chemistry Letters 2016, 26, 730\u0026ndash;733.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Kim, H. K.; Choi, S. U.; Moon, E.; Kim, S. Y.; Lee, K. R., Bioactive lignans from the rhizomes of Acorus gramineus. Journal of Natural Products 2011, 74, 2187\u0026ndash;2192.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Kwon, O. W.; Kim, S. Y.; Lee, K. R., Bioactive lignans from the trunk of Abies holophylla. Journal of Natural Products 2013, 76, 2131\u0026ndash;2135.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Ha, S. K.; Choi, S. U.; Kim, S. Y.; Lee, K. R., Bioactive phenolic constituents from the seeds of Pharbitis nil. Chemical Pharmaceutical Bulletin 2011, 59, 1425\u0026ndash;1429.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Moon, E.; Kim, S. Y.; Choi, S. U.; Son, M. W.; Choi, S. Z.; Lee, K. R., Bioactive sesquiterpenes from the essential oil of Thuja orientalis. Planta Medica 2013, 79, 1680\u0026ndash;1684.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Subedi, L.; Oh, J.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Bioactive Triterpenoids from the Twigs of Chaenomeles sinensis. Journal of Natural Products 2017, 80, 1134\u0026ndash;1140.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark, K. J.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Bioactive triterpenoids from twigs of Betula schmidtii. Bioorganic Chemistry 2018, 77, 527\u0026ndash;533.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Moon, E.; Choi, S. U.; Kim, S. Y.; Lee, K. R., Biological evaluation of phenolic constituents from the trunk of Berberis koreana. Bioorganic Medicinal Chemistry Letters 2011, 21, 2270\u0026ndash;2273.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuh, W. S.; Park, K. J.; Kim, D. H.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., A biphenyl derivative from the twigs of Chaenomeles speciosa. Bioorganic Chemistry 2017, 72, 156\u0026ndash;160.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Choi, S. U.; Ha, S. K.; Kim, S. Y.; Lee, K. R., Biphenyls from Berberis koreana. Journal of Natural Products 2009, 72, 2061\u0026ndash;2064.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, D. H.; Lee, T. H.; Subedi, L.; Kim, S. Y.; Lee, K. R., Chemical constituents of Impatiens balsamina stems and their biological activities. Natural Product Sciences 2019, 25, 130\u0026ndash;135.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Choi, J. W.; Choi, S. U.; Ha, S. K.; Kim, S. Y.; Park, H.-J.; Lee, K. R., The chemical constituents of Piper kadsura and their cytotoxic and anti-neuroinflammtaory activities. J Enzym Inhib Med Ch 2011, 26, 254\u0026ndash;260.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSo, H. M.; Yu, J. S.; Khan, Z.; Subedi, L.; Ko, Y.-J.; Lee, I. K.; Park, W. S.; Chung, S. J.; Ahn, M.-J.; Kim, S. Y., Chemical constituents of the root bark of Ulmus davidiana var. japonica and their potential biological activities. Bioorganic Chemistry 2019, 91, 103145.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Moon, E.; Lee, S. R.; Park, K. J.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Chemical Constituents of the Seeds of Raphanus sativus and their Biological Activity. Journal of the Brazilian Chemical Society 2015, 26, 2307\u0026ndash;2312.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVenkanna, A.; Cho, K. H.; Dhorma, L. P.; Kumar, D. N.; Hah, J. M.; Park, H.-g.; Kim, S. Y.; Kim, M.-h., Chemistry-oriented synthesis (ChOS) and target deconvolution on neuroprotective effect of a novel scaffold, oxaza spiroquinone. \u003cem\u003eEuropean Journal of Medicinal Chemistry\u003c/em\u003e 2019, 163, 453\u0026ndash;480.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHyun Kim, K.; Keun Ha, S.; Yeou Kim, S.; Joo Youn, H.; Ro Lee, K., Constituents of Limonia acidissima inhibit LPS-induced nitric oxide production in BV-2 microglia. Journal of enzyme inhibition medicinal chemistry 2010, 25, 887\u0026ndash;892.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Kim, K. H.; Lee, K. R., Diterpenes from the Trunk of Abies holophylla and Their Potential Neuroprotective and Anti-inflammatory Activities. Journal of Natural Products 2016, 79, 387\u0026ndash;394.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWoo, K. W.; Moon, E.; Park, S. Y.; Kim, S. Y.; Lee, K. R., Flavonoid glycosides from the leaves of Allium victorialis var. platyphyllum and their anti-neuroinflammatory effects. Bioorg Med Chem Lett 2012, 22, 7465\u0026ndash;7470.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Oh, J.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Holophyllane A: A Triterpenoid Possessing an Unprecedented B-nor-3,4-seco-17,14-friedo-lanostane Architecture from Abies holophylla. Scientific Reports 2017, 7, 1\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Shin, B.; Kwon, O. W.; Kim, S. Y.; Choi, S. U.; Oh, D.-C.; Kim, K. H.; Lee, K. R., Holophyllin A, a rearranged abietane-type diterpenoid from the trunk of Abies holophylla. Tetrahedron Letters 2014, 55, 6504\u0026ndash;6507.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Oh, J.; Suh, W. S.; Jang, S. W.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Investigation of chemical constituents from Spiraea prunifolia var. simpliciflora and their biological activities. Phytochemistry Letters 2017, 22, 255\u0026ndash;260.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuh, W. S.; Kim, C. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Iridoid Glycosides from the Twigs of Sambucus williamsii var. coreana and Their Biological Activities. Journal of Natural Products 2017, 80, 2502\u0026ndash;2508.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Subedi, L.; Kwon, O. K.; Kim, S. Y.; Yeo, E.-J.; Choi, S. U.; Lee, K. R., Isolation of bioactive biphenyl compounds from the twigs of Chaenomeles sinensis. Bioorganic Medicinal Chemistry Letters 2016, 26, 351\u0026ndash;354.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Moon, E.; Choi, S. U.; Kim, S. Y.; Lee, K. R., Lanostane triterpenoids from the mushroom Naematoloma fasciculare. Journal of Natural Products 2013, 76, 845\u0026ndash;851.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Kim, K. H.; Lee, K. R., Lignan Glycosides from the Twigs of Chaenomeles sinensis and Their Biological Activities. Journal of Natural Products 2015, 78, 1174\u0026ndash;1178.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K.-H.; Ha, S.-K.; Kim, S.-Y.; Kim, S.-H.; Lee, K.-R., Limodissimin A: A New Dimeric Coumarin from Limonia acidissima. Bulletin of the Korean Chemical Society 2009, 30, 2135\u0026ndash;2137.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDung, H. V.; Bach, N. V.; Trung, T. N.; Nhiem, N. X.; Tai, B. H.; Kiem, P. V.; Park, S.; Lee, T. H.; Kim, S. Y.; Kim, S. H., Megastigmane glycosides from Docynia indica and their anti-inflammatory activities. Helvetica Chimica Acta 2016, 99, 681\u0026ndash;686.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSubedi, L.; Kwon, O. W.; Pak, C.; Lee, G.; Lee, K.; Kim, H.; Kim, S. Y., N,N-disubstituted azines attenuate LPS-mediated neuroinflammation in microglia and neuronal apoptosis via inhibiting MAPK signaling pathways. BMC Neuroscience 2017, 18, 1\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Choi, J. W.; Ha, S. K.; Kim, S. Y.; Lee, K. R., Neolignans from Piper kadsura and their anti-neuroinflammatory activity. Bioorganic medicinal chemistry letters 2010, 20, 409\u0026ndash;412.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Suh, W. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Neuroprotective Fatty Acids from the Stem Bark of Sorbus commixta. Lipids 2016, 51, 989\u0026ndash;995.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan Kiem, P.; Cuong, L. C. V.; Trang, D. T.; Nhiem, N. X.; Anh, H. L. T.; Tai, B. H.; Huong, L. M.; Van Minh, C.; Lee, T. H.; Kim, S. Y., New Alkaloids and Anti-inflammatory Constituents from the Leaves of Antidesma ghaesembilla. Natural Product Communications 2017, 12, 11\u0026ndash;14.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, T. H.; Khan, Z.; Subedi, L.; Kim, S. Y.; Lee, K. R., New bis-thioglycosyl-1, 1\u0026prime;-disulfides from Nasturtium officinale R. Br. and their anti-neuroinflammatory effect. Bioorganic chemistry 2019, 86, 501\u0026ndash;506.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Kwon, O. W.; Kim, S. Y.; Kim, K. H.; Lee, K. R., A New Cyclic Triterpene Saponin from Phyteuma Japonicum. Heterocycles 2014, 89, 1913\u0026ndash;1922.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Subedi, L.; Oh, J.; Kim, S. Y.; Choi, S. U.; Lee, K. R., A New Phenolic Compound from Salix Glandulosa. Heterocycles 2018, 96, 935\u0026ndash;946.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, J. S.; Baek, J.; Park, H. B.; Moon, E.; Kim, S. Y.; Choi, S. U.; Kim, K. H., A new rearranged eudesmane sesquiterpene and bioactive sesquiterpenes from the twigs of Lindera glauca (Sieb. et Zucc.) Blume. Archives of Pharmacal Research 2016, 39, 1628\u0026ndash;1634.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Lee, I. K.; Choi, S. U.; Lee, J. H.; Moon, E.; Kim, S. Y.; Lee, K. R., New triterpenoids from the tubers of Corydalis ternata: structural elucidation and bioactivity evaluation. Planta Medica 2011, 77, 1555\u0026ndash;1558.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. H.; Moon, E.; Kim, H. K.; Oh, J. Y.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Phenolic constituents from the rhizomes of Acorus gramineus and their biological evaluation on antitumor and anti-inflammatory activities. Bioorg Med Chem Lett 2012, 22, 6155\u0026ndash;6159.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark, K. J.; Cha, J. M.; Subedi, L.; Kim, S. Y.; Lee, K. R., Phenolic constituents from the twigs of Betula schmidtii collected in Goesan, Korea. Phytochemistry 2019, 167, 112085.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWoo, K. W.; Suh, W. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Kim, K. H.; Lee, K. R., Phenolic Derivatives from the Stems of Lagerstroemia Indica and Their Biological Activity. Heterocycles 2015, 91, 2355\u0026ndash;2366.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Kwon, O. W.; Kim, S. Y.; Choi, S. U.; Kim, J. Y.; Han, J. Y.; Choi, S. I.; Choi, J. G.; Kim, K. H.; Lee, K. R., Phenolic Glycosides from the Twigs of Salix glandulosa. Journal of Natural Products 2014, 77, 1955\u0026ndash;1961.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCha, J. M.; Kim, D. H.; Lee, T. H.; Subedi, L.; Kim, S. Y.; Lee, K. R., Phytochemical constituents of Capsella bursa-pastoris and their anti-inflammatory activity. Natural Product Sciences 2018, 24, 132\u0026ndash;138.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, S. Y.; Moon, E.; Kim, S. Y.; Lee, K. R., Quinic acid derivatives from Pimpinella brachycarpa exert anti-neuroinflammatory activity in lipopolysaccharide-induced microglia. Bioorganic Medicinal Chemistry Letters 2013, 23, 2140\u0026ndash;2144.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Oh, J.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Rare Thioglycosides from the Roots of Wasabia japonica. Journal of Natural Products 2018, 81, 2129\u0026ndash;2133.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuh, W. S.; Kwon, O. K.; Lee, T. H.; Subedi, L.; Kim, S. Y.; Lee, K. R., Secoiridoid Glycosides from the Twigs of Ligustrum obtusifolium Possess Anti-inflammatory and Neuroprotective Effects. Chemical Pharmaceutical Bulletin 2018, 66, 78\u0026ndash;83.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark, K. J.; Kim, C. S.; Khan, Z.; Oh, J.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Securinega Alkaloids from the Twigs of Securinega suffruticosa and Their Biological Activities. Journal of Natural Products 2019, 82, 1345\u0026ndash;1353.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan Kiem, P.; Hang, D. T.; Nhiem, N. X.; Tai, B. H.; Anh, H. L. T.; Van Cuong, P.; Quang, T. H.; Van Minh, C.; Van Dau, N.; Kim, Y.-A.; Subedi, L.; Kim, S. Y.; Kim, S. H., Sesquiterpene derivatives from marine sponge Smenospongia cerebriformis and their anti-inflammatory activity. Bioorganic medicinal chemistry letters 2017, 27, 1525\u0026ndash;1529.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Kim, S. Y.; Moon, E.; Lee, M. K.; Lee, K. R., Steroidal constituents from the leaves of Hosta longipes and their inhibitory effects on nitric oxide production. Bioorganic Medicinal Chemistry Letters 2013, 23, 1771\u0026ndash;1775.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Oh, J.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Lee, K. R., Structural Characterization of Terpenoids from Abies holophylla Using Computational and Statistical Methods and Their Biological Activities. Journal of Natural Products 2018, 81, 1795\u0026ndash;1802.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYuan, Y.; Subedi, L.; Lim, D.; Jung, J.-K.; Kim, S. Y.; Seo, S.-Y., Synthesis and anti-neuroinflammatory activity of N-heterocyclic analogs based on natural biphenyl-neolignan honokiol. Bioorganic Medicinal Chemistry Letters 2019, 29, 329\u0026ndash;333.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShobha, D.; Chari, M. A.; Mukkanti, K.; Kim, S. Y., Synthesis and anti-neuroinflammatory activity studies of substituted 3,4-dihydroquinoxalin-2-amine derivatives. Tetrahedron Letters 2012, 53, 2675\u0026ndash;2679.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJin, G. H.; Ha, S. K.; Park, H. M.; Kang, B.; Kim, S. Y.; Kim, H.-D.; Ryu, J.-H.; Jeon, R., Synthesis of azaisoflavones and their inhibitory activities of NO production in activated microglia. Bioorg Med Chem Lett 2008, 18, 4092\u0026ndash;4094.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, T. H.; Khan, Z.; Kim, S. Y.; Lee, K. R., Thiohydantoin and Hydantoin Derivatives from the Roots of Armoracia rusticana and Their Neurotrophic and Anti-neuroinflammatory Activities. Journal of Natural Products 2019, 82, 3020\u0026ndash;3024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWoo, K. W.; Park, J. E.; Cha, J. M.; Subedi, L.; Kim, S. Y.; Lee, K. R., Three New Lignan Glycosides from the Firmiana simplex. Chemical Pharmaceutical Bulletin 2019, 67, 18\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, C. S.; Subedi, L.; Kim, S. Y.; Choi, S. U.; Choi, S. Z.; Son, M. W.; Kim, K. H.; Lee, K. R., Two new phenolic compounds from the white flower of Impatiens balsamina. Phytochemistry Letters 2015, 14, 215\u0026ndash;220.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuh, W. S.; Lee, S. Y.; Park, J. E.; Kim, D. H.; Kim, S. Y.; Lee, K. R., Two New Steroidal Alkaloids from the Bulbs of Fritillaria Thunbergii. Heterocycles 2018, 96, 925\u0026ndash;934.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTuan Anh, H. L.; Kim, D.-C.; Ko, W.; Ha, T. M.; Nhiem, N. X.; Yen, P. H.; Tai, B. H.; Truong, L. H.; Long, V. N.; Gioi, T. J. P. b., Anti-inflammatory coumarins from Paramignya trimera. 2017, 55, 1195\u0026ndash;1201.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYoon, C.-S.; Kim, K.-W.; Lee, S.-C.; Kim, Y.-C.; Oh, H. J. B.; letters, m. c., Anti-neuroinflammatory effects of sesquiterpenoids isolated from Nardostachys jatamansi. 2018, 28, 140\u0026ndash;144.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, N.; Wang, Y.; Li, X.; Zhang, H.; Zhou, D.; Wang, W.; Li, W.; Zhang, X.; Li, X.; Hou, Y. J. B.; Letters, M. C., Bioactive phenols as potential neuroinflammation inhibitors from the leaves of Xanthoceras sorbifolia Bunge. 2016, 26, 5018\u0026ndash;5023.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, G.; Xie, Y.; Zhou, D.; Yang, Y.; Liu, J.; Hou, Y.; Cheng, M.; Liu, Y.; Li, N. J. P., Chemical constituents from shells of Xanthoceras sorbifolium. 2020, 172, 112288.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, S. R.; Kim, M. S.; Kim, S.; Hwang, K. W.; Park, S. Y. J. C.; Biodiversity, Constituents from Scutellaria barbata inhibiting nitric oxide production in LPS-stimulated microglial cells. 2017, 14, e1700231.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, D.-C.; Quang, T. H.; Ngan, N. T. T.; Yoon, C.-S.; Sohn, J. H.; Yim, J. H.; Feng, Y.; Che, Y.; Kim, Y.-C.; Oh, H. J. J. o. n. p., Dihydroisocoumarin derivatives from marine-derived fungal isolates and their anti-inflammatory effects in lipopolysaccharide-induced BV2 microglia. 2015, 78, 2948\u0026ndash;2955.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOck, J.; Han, H. S.; Hong, S. H.; Lee, S. Y.; Han, Y. M.; Kwon, B. M.; Suk, K. J. B. j. o. p., Obovatol attenuates microglia-mediated neuroinflammation by modulating redox regulation. 2010, 159, 1646\u0026ndash;1662.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYeh, C. H.; Yang, M. L.; Lee, C. Y.; Yang, C. P.; Li, Y. C.; Chen, C. J.; Kuan, Y. H. J. E. t., Wogonin attenuates endotoxin-induced prostaglandin E2 and nitric oxide production via Src‐ERK1/2‐NFκB pathway in BV‐2 microglial cells. 2014, 29, 1162\u0026ndash;1170.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K.-W.; Quang, T. H.; Ko, W.; Kim, D.-C.; Yoon, C.-S.; Oh, H.; Kim, Y.-C. J. P. b., Anti-neuroinflammatory effects of cudraflavanone A isolated from the chloroform fraction of Cudrania tricuspidata root bark. 2018, 56, 192\u0026ndash;200.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchr\u0026ouml;dinger Maestro, Schr\u0026ouml;dinger, LLC, New York, 2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAxen, S. D.; Huang, X.-P.; C\u0026aacute;ceres, E. L.; Gendelev, L.; Roth, B. L.; Keiser, M. J., A Simple Representation of Three-Dimensional Molecular Structure. Journal of Medicinal Chemistry 2017, 60, 7393\u0026ndash;7409.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuan, J.; Dixon, S. L.; Lowrie, J. F.; Sherman, W., Analysis and comparison of 2D fingerprints: insights into database screening performance using eight fingerprint methods. J Mol Graph Model 2010, 29, 157\u0026ndash;170.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSastry, M.; Lowrie, J. F.; Dixon, S. L.; Sherman, W., Large-scale systematic analysis of 2D fingerprint methods and parameters to improve virtual screening enrichments. Journal of Chemical Information Modeling 2010, 50, 771\u0026ndash;784.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchr\u0026ouml;dinger Canvas, Schr\u0026ouml;dinger, LLC, New York, NY, 2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchr\u0026ouml;dinger LigPrep, Schr\u0026ouml;dinger, LLC, New York, 2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchr\u0026ouml;dinger ConfGen, Schr\u0026ouml;dinger, LLC, New York, 2019.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWatts, K. S.; Dalal, P.; Murphy, R. B.; Sherman, W.; Friesner, R. A.; Shelley, J. C., ConfGen: a conformational search method for efficient generation of bioactive conformers. Journal of Chemical Information Modeling 2010, 50, 534\u0026ndash;546.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHawkins, P. C.; Skillman, A. G.; Warren, G. L.; Ellingson, B. A.; Stahl, M. T., Conformer generation with OMEGA: algorithm and validation using high quality structures from the Protein Databank and Cambridge Structural Database. Journal of Chemical Information Modeling 2010, 50, 572\u0026ndash;584.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOpenEye \u003cem\u003eOMEGA 3.0.1.2\u003c/em\u003e, OpenEye Scientific Software, Santa Fe, NM \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.eyesopen.com\u003c/span\u003e\u003cspan address=\"http://www.eyesopen.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e., 2018.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHemmerich, J.; Asilar, E.; Ecker, G. F., COVER: conformational oversampling as data augmentation for molecules. Journal of Cheminformatics 2020, 12, 1\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLama, P. D.; Kumar, S.; Kim, K.; Ahn, S.; Kim, M. H., Quantitative Prediction on the Enantioselectivity of Multiple Chiral Iodoarene Scaffolds Based on Whole Geometry. 2021, arXiv preprint arXiv:2103.14065.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbadi, M.; Barham, P.; Chen, J.; Chen, Z.; Davis, A.; Dean, J.; Devin, M.; Ghemawat, S.; Irving, G.; Isard, M. Tensorflow: A system for large-scale machine learning. In 12th {USENIX} symposium on operating systems design and implementation ({OSDI} 16), 2016; 2016; pp\u0026nbsp;265\u0026ndash;283.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V. J. t. J. o. m. L. r., Scikit-learn: Machine learning in Python. 2011, 12, 2825\u0026ndash;2830.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTaqi, A. M.; Awad, A.; Al-Azzo, F.; Milanova, M., The impact of multi-optimizers and data augmentation on TensorFlow convolutional neural network performance. In 2018 \u003cem\u003eIEEE Conference on Multimedia Information Processing and Retrieval (MIPR)\u003c/em\u003e 2018, 140.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlgamal, Z. Y.; Lee, M. H., Penalized logistic regression with the adaptive LASSO for gene selection in high-dimensional cancer classification. Expert Systems with Applications 2015, 42, 9326\u0026ndash;9332.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFeng, Z. Z.; Yang, X.; Subedi, S.; McNicholas, P. D., The LASSO and sparse least squares regression methods for SNP selection in predicting quantitative traits. IEEE/ACM transactions on computational biology 2011, 9, 629\u0026ndash;636.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZou, H.; Hastie, T., Regularization and variable selection via the elastic net. Journal of the royal statistical society: series B 2005, 67, 301\u0026ndash;320.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu, S. Bayesian Na\u0026iuml;ve Bayes classifiers to text classification. Journal of Information Science, 2018, 44(1), 48\u0026ndash;59.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChicco, D.; Jurman, G., The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation. BMC Genomics 2020, 21, 6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMatthews, B. W. J. B. e. B. A.-P. S., Comparison of the predicted and observed secondary structure of T4 phage lysozyme. 1975, 405, 442\u0026ndash;451.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePowers, D. M., Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation. arXiv preprint 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAraujo, M. B., Pearson, R. G., Thuiller, W., \u0026amp; Erhard, M. (2005). Validation of species\u0026ndash;climate impact models under climate change. Global change biology, 11(9), 1504\u0026ndash;1513.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, S. H.; Ahn, S.; Kim, M. H., Comparing a query compound with drug target classes using 3D-chemical similarity. International Journal of Molecular Sciences, 2020, 21(12), 4208.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhn, S.; Lee, S. E.; Kim, M. H.. Random-forest model for drug\u0026ndash;target interaction prediction via Kullback\u0026ndash;Leibler divergence. Journal of Cheminformatics, 2022, 14(1), 1\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAizawa, A., An information-theoretic perspective of tf\u0026ndash;idf measures. Information Processing \u0026amp; Management, 2003, 39(1), 45\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Molecular Featurization, Phenotype Structure Relationship, Neuroinflammation, Microglia Activation, Location Information","lastPublishedDoi":"10.21203/rs.3.rs-3812369/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3812369/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cem\u003eCNS Drug discovery has been challenging due to the lack of clarity on CNS diseases' basic biological and pathological mechanisms. Despite the difficulty, some CNS drugs have been developed based on phenotypic effects. Herein, we propose a phenotype-structure relationship model, which predicts an anti-neuroinflammatory potency based on 3D molecular structures of the phenotype-active or inactive compounds without specifying targets. For this chemo-centric study, a predictive model of the nitric oxide (NO) inhibitory potency in hyper-activated microglia is built from the 548 agents, which were collected from \u003c/em\u003e95 research articles \u003cem\u003e(28 substructures consisting of natural products and synthetic scaffolds) and doubly externally validated by the agents of \u003c/em\u003e9 research articles\u003cem\u003e as third set. 3D Structures (multi-conformer ensemble) of every agent were encoded into the E3FP molecular fingerprint of the Keiser group as a 3D molecular representation. The location information of the molecular fingerprints could be learned and validated to classify the inhibitory potency of compounds (IC\u003c/em\u003e\u003csub\u003e\u003cem\u003e50 \u003c/em\u003e\u003c/sub\u003e\u003cem\u003ecut-off between bi-classes: 37.1 μM): (1) multi-layer perceptron (MLP) (accuracy: 0.962, AUC: 0.994), (2) recurrent neural network (RNN) (accuracy: 0.966, AUC: 0.994), and (3) convolutional neural network (CNN) (accuracy: 0.969, AUC: 0.996). The high performance of these models was compared with that of four classical machine classification models (Logistic, Ridge, Lasso, and Naïve Bayes). \u003c/em\u003eWe named the bi-class models NO-Classifier. \u003cem\u003eOut-of-set validation and decision region analysis of the out-of-set doubly demonstrated \u003c/em\u003eNO-Classifier\u003cem\u003e effectively discerned the anti-inflammatory potency of testing compounds in inflammatory cell phenotype with the webserver in https://no-classifier.onrender.com.\u003c/em\u003e\u003c/p\u003e","manuscriptTitle":"NO-Classifier: Prediction of Anti- Neuroinflammatory Agents Using Text Mining of 3D Molecular Fingerprints","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-01 08:09:01","doi":"10.21203/rs.3.rs-3812369/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-07-02T04:51:53+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-06-30T18:00:42+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-06-25T01:44:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"13170395611225346465826726553511756165","date":"2024-06-23T00:28:25+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"265064516304352747500496450628255573367","date":"2024-06-21T16:25:03+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-04-08T06:45:59+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"89c48678-2b2b-4152-a028-45ff671b6ddf","date":"2024-03-28T16:02:05+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-03-28T12:51:00+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-02-13T12:16:44+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2023-12-28T11:28:31+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2023-12-28T11:26:46+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2023-12-27T13:08:29+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"14fadae0-aded-403d-b218-5aa80f611513","owner":[],"postedDate":"January 1st, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":27879209,"name":"Biological sciences/Computational biology and bioinformatics/Computational models"},{"id":27879210,"name":"Biological sciences/Computational biology and bioinformatics/Data processing"},{"id":27879211,"name":"Biological sciences/Computational biology and bioinformatics/High throughput screening"},{"id":27879212,"name":"Biological sciences/Computational biology and bioinformatics/Machine learning"},{"id":27879213,"name":"Biological sciences/Drug discovery/Drug screening"},{"id":27879214,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":27879215,"name":"Biological sciences/Drug discovery"},{"id":27879216,"name":"Health sciences/Molecular medicine"},{"id":27879217,"name":"Physical sciences/Chemistry"},{"id":27879218,"name":"Physical sciences/Chemistry/Cheminformatics"}],"tags":[],"updatedAt":"2024-11-18T19:19:16+00:00","versionOfRecord":{"articleIdentity":"rs-3812369","link":"https://doi.org/10.1038/s41598-024-78823-3","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2024-11-16 15:58:15","publishedOnDateReadable":"November 16th, 2024"},"versionCreatedAt":"2024-01-01 08:09:01","video":"","vorDoi":"10.1038/s41598-024-78823-3","vorDoiUrl":"https://doi.org/10.1038/s41598-024-78823-3","workflowStages":[]},"version":"v1","identity":"rs-3812369","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3812369","identity":"rs-3812369","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}