DCPM-ADMET: Fusion of Dual-channel Pre-trained Model and Molecular Fingerprints to enhance Drug ADMET Properties Prediction

DCPM-ADMET: Fusion of Dual-channel Pre-trained Model and Molecular Fingerprints to enhance Drug ADMET Properties Prediction | 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 Research Article DCPM-ADMET: Fusion of Dual-channel Pre-trained Model and Molecular Fingerprints to enhance Drug ADMET Properties Prediction Yuchen Zeng, Yue Qi, Leilei Zhang, Kaili Jiang, Xiaofei Zhou, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8326331/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract The absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties of drugs are critical to their efficacy and safety in clinical trials; however, traditional machine learning methods have limited generalization ability in ADMET prediction due to insufficient data. To address this issue, we developed DCPM-ADMET, an innovative pre-trained model with higher accuracy, whose architecture employs a two-channel system—including an XLNet-based module for capturing the semantics of molecular sequences, an RNN-based component for small molecule property extraction, and ECFP fingerprints for capturing molecular substructures—and after initial pre-training, the model outperforms traditional methods in prediction accuracy on multiple benchmark datasets for molecular properties; additionally, we fine-tuned it on a self-constructed database containing 465,470 entries covering 97 ADMET properties, and by integrating these 97 prediction models and 36 computational properties, we further developed a free online ADMET prediction tool with 133 endpoints (available at http://admet.bioai-global.com/ ), which is designed to assist researchers in conducting comprehensive molecular ADMET predictions. Scientific contribution The development of DCPM-ADMET represents a seminal advancement in computational pharmacology. This novel pre-trained model successfully addresses the fundamental limitation of poor generalization in predicting ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties, a challenge stemming from data insufficiency in traditional machine learning approaches. Our architecture innovatively employs a dual-channel system: an XLNet-based module for deep capture of molecular sequence semantics, an RNN-based component for efficient extraction of small molecule properties, and ECFP fingerprints to comprehensively encode structural features. Following intensive pre-training, DCPM-ADMET demonstrates superior predictive accuracy across multiple benchmark molecular property datasets. Furthermore, we fine-tuned this model on a proprietary, large-scale database of 465,470 entries covering 97 ADMET endpoints. By integrating the resultant 97 prediction models with 36 calculated physicochemical properties, we have deployed a free, high-throughput online ADMET prediction tool with 133 endpoints, which is set to become an essential resource for guiding early-stage drug discovery and safety assessment. Pretraining ADMET XLNet RNN Molecular Fingerprints Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Drug development is a complex, time-consuming, and expensive process. Traditional drug development methods often require years and substantial financial investment to bring a new drug to market[ 1 ]. The pivotal success of clinical trials significantly shapes the course of drug development, holding paramount importance for pharmaceutical investors and patients alike. Nevertheless, about 40–60% of drug candidates succumb to unfavorable ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) characteristics during clinical trials[ 2 ]. By anticipating these ADMET properties, researchers can pinpoint and eradicate unfavorable pharmacokinetic traits in prospective compounds, expediting the developmental process while curbing time and financial expenses. This proactive strategy also facilitates the early recognition of potentially efficacious compounds during screening, prioritizing them for subsequent advancements and refining the selection of candidates with the highest likelihood of success as novel medications. Furthermore, it underpins the determination of optimal dosing, optimal routes of administration, and minimizes the threat of adverse reactions, enhancing both the safety and efficacy of drug candidates[ 3 ][ 4 ]. In recent years, the landscape of computational simulation software for ADMET property prediction has undergone remarkable advancements, achieving notable strides in enhancing prediction accuracy[ 5 ]. To further refine this intricate endeavor, artificial intelligence technologies have emerged as pivotal players in diverse ADMET computational frameworks, yielding an array of outstanding prediction tools, including ADMETlab, admetSAR, Interpretable-ADMET, HelixADMET, among others[ 6 ][ 7 ][ 8 ][ 9 ][ 10 ]. Nevertheless, the application of current models confronts myriad challenges, encompassing poor data quality, limited generalization capabilities, interpretability gaps, data scarcity and uncertainties, thereby restricting model accuracy and reliability. These hurdles underscore the urgency for continued research and optimization. The algorithmic evolution in drug ADMET prediction has witnessed profound transitions, evolving from traditional machine learning through deep learning to the advent of pre-trained models. Initially, researchers harnessed traditional methods like linear regression, support vector machines, and random forests, relying on manual feature engineering to capture molecular attributes[ 11 ]. However, these approaches often grappled with the intricacies of ADMET prediction. As computational prowess and data availability soared, deep learning techniques emerged, automating feature learning and demonstrating exceptional performance in ADMET prediction tasks[ 12 ]. Nevertheless, their hunger for massive datasets and challenges in generalization amidst data scarcity remain obstacles. Recently, pre-trained models have emerged as promising avenues for ADMET prediction. By pre-training on extensive datasets, these models acquire rich feature representations that are subsequently fine-tuned for specific ADMET tasks, significantly bolstering generalization and prediction precision. They not only adapt seamlessly to diverse ADMET prediction scenarios but also address data scarcity and uncertainty more adeptly, marking the cutting edge of algorithmic developments in this field[ 13 ]. In the present work, the model underwent an extensive unsupervised pre-training regimen utilizing a vast small molecule dataset. The pre-training employs a dual-channel architecture to discern intricate and abstract feature representations from different aspects, which we refer to as Dual-channel Pre-trained Model(DCPM). One channel, based on the XLNet architecture, is used to learn the semantic differences in molecular sequences, while the other, utilizing an RNN-based architecture, captures complex representations at the level of molecular properties. Subsequently, DCPM was trained and evaluated across a broad spectrum of downstream tasks, encompassing various challenges within the MoleculeNet benchmark dataset, such as predicting Caco-2 Permeability, BBB Penetration, and identifying CYP450 1A2 Inhibitors. Its state-of-the-art performance and robust generalization capabilities across these tasks validated its accuracy and reliability. Building upon this success, we further extended DCPM's reach by subjecting it to rigorous training and testing within 97 downstream ADMET tasks. We conducted a comparison with the traditional molecular fingerprint ECFP and our method demonstrated superiority in nearly 70% of the tasks. In addition, when compared with existing tools, we encompass the largest number of ADMET regression tasks designed to deliver more precise prediction outcomes, and simultaneously, our overall dataset is also the largest, containing over 460 000 entries, thereby laying a solid foundation for a more holistic and comprehensive approach to drug ADMET prediction. Materials and Methods Data Preparation Pre-training data The DCPM-ADMET training process incorporates diverse datasets, encompassing chemical molecule collections for pre-training and targeted ADMET datasets for constructing predictive models. The pre-training phase leverages PubChem[14], a comprehensive bioactivity database housing millions of compound records spanning pharmaceuticals, natural products, and chemicals. From this repository, we extracted SMILES and InChI representations of chemical molecules, subjecting them to a rigorous data preparation pipeline: 1. Data Standardization: Utilizing RDKit (version 2023.9.6), SMILES strings underwent standardization procedures, including bond order canonicalization and removal of redundant hydrogen atoms. 2. Missing & Duplicate Value Filtering: Molecules that failed standardization were deemed missing values, the missing and duplicate values are removed to ensure data uniqueness and validity. 3. Descriptor Calculation: Based on the standardized SMILES, we employed RDKit to compute nine crucial physicochemical properties for each molecule, serving as molecular descriptors. These encompassed logP, maximal and minimal partial charges, valence electron count, hydrogen bond donor and acceptor numbers, Balaban's J value, molar refractivity, and topological polar surface area. These processes yielded 111 295 885 data points for pre-training. ADMET datasets For the ADMET property datasets, we aggregated resources from multiple experimental compound databases, notably PubChem, ChEMBL[15] and TDC[16], and implemented a meticulous data processing strategy: 1. Data Standardization: SMILES representations were standardized using RDKit, and harmonizing data across datasets into consistent units like pEC50 and mol/L. 2. Data Cleaning & Merging: Rigorous comparison of entries across similar datasets from different sources led to the removal of conflicting samples in classification sets and unreliable samples in regression sets exhibiting discrepancies exceeding an order of magnitude. Subsequently, datasets were merged and further cleaned. 3. Duplicate Data Elimination: Following cleaning and merging, duplicate entries were purged, resulting in a refined ADMET dataset. This final dataset comprises 97 ADMET sets, segmented into 43 regression and 54 classification datasets, totaling 465 470 data entries. Model Architecture The DCPM-ADMET molecular property prediction framework encapsulates two essential components: a feature extraction module and a downstream task prediction module. The feature extraction module meticulously crafts deep and versatile vector representations from SMILES molecular sequences. These representations then underpin tailored modeling for diverse downstream tasks. Pre-training Inspired by Transformer-based models' exceptional performance, DCPM embraces pre-trained architectures to refine SMILES sequences. However, prevalent methods like MoLFormer[17] and KPGT[18], which adopt the Transformer encoder during pre-training with masked token recovery akin to BERT[19], confront a pre-training-fine-tuning disparity when real-world inputs lack mask tokens[20]. 1. XLNet[21], innovatively combining autoregressive and permutation strategies, resolves this inconsistency while enabling bidirectional context analysis, a feat previously unattainable by autoregressive models. By leveraging XLNet's permutation technique and dual-stream attention, DCPM captures global SMILES sequence patterns through self-attention and distills chemical insights from bidirectionally permuted contexts, enriching feature representation. 2. In the realm of Natural Language Processing (NLP), XLNet leverages the Transformer-XL architecture to enhance its capability in managing extremely long texts. Nevertheless, the SMILES sequences of chemical molecules, characterized by their relatively shorter lengths and smaller vocabularies, are more susceptible to overfitting challenges. To mitigate this, we have omitted the long text processing feature and drawing incorporated an additional RNN structure inspiration from the approach of Robin Winter[12]. This RNN conducts a Seq2Seq SMILES-to-InChI translation while regressing nine physicochemical descriptors. By integrating InChI's fundamental physicochemical insights and descriptor regression into the pre-training objectives, we augment task complexity, mitigating overfitting, and fostering deeper molecular knowledge extraction beyond mere sequence reconstruction. 3. Considering that tokenized SMILES sequences mainly consist of atomic tokens, lacking substructure fragment information of chemical molecules, we added a feature fusion layer to DCPM. This layer amalgamates XLNet and RNN hidden layers with Extended Connectivity Fingerprints (ECFP)[22], an atomic connectivity graph-based descriptor rich in topological and substructure information, augmenting the model's substructural comprehension. In Fig. 1a, the pre-training workflow is outlined: SMILES are tokenized, embedded, and processed by an Encoder amalgamating GRU cells and XLNet layers. Extracted features are fed into three decoders, optimizing across InChI reconstruction, property prediction, and masked structure tasks. Fine-tuning For fine-tuning, depicted in Fig. 1b, the feature extraction layer is frozen, while the downstream prediction layer is trained. For binary classification and regression tasks, we employ Random Forest to predict properties. The deep neural network applied in multi-task uses three hidden layers, optimizing key hyperparameters such as L2 regularization, dropout, and the number of neurons in each hidden layer. The Tree-structured Parzen Estimator(TPE) algorithm is used to identify the best hyperparameters across 50 evaluations. Early stopping and a random 8:1:1 train-validate-test split with AUC metrics over five runs ensure robustness. DCPM's prowess is validated against ECFP on MoleculeNet, demonstrating superiority across 10 subsets (7 classification and 3 regression tasks). Among the seven classification datasets, ClinTox, SIDER, Tox21, and ToxCast are multi-label classification task, while the other three classification datasets (BACE, BBBP, and HIV) and the regression datasets (ESOL, Lipophilicity, FreeSolv) are single-task. Ultimately, DCPM-ADMET is trained on a comprehensive ADMET dataset, encompassing 54 classification and 43 regression tasks, solidifying its standing as a potent tool for ADMET property prediction. Directly computable properties For enhanced usability, we have integrated 36 additional directly computable endpoints (see the Supplementary Information calculated_props for detailed values), encompassing the calculation of physicochemical properties, medicinal chemistry principles, and toxicophore rules using the RDKit and Scopy packages[23]. Results and Discussion Results and Implementation of DCPM-ADMET First, when compared with traditional fingerprint features, ECFP, across 10 benchmark datasets, DCPM outperformed traditional fingerprints in 7 out of 10 tasks, demonstrating a clear advantage (see Table 1 and Table 2 ). Table 1 Comparison of DCPM and fingerprint on classification task Method Classification (ROC-AUC) BACE BBBP HIV ClinTox SIDER Tox21 ToxCast ecfp 0.888 (0.023) 0.909 (0.011) 0.811 (0.015) 0.978 (0.009) 0.815 (0.006) 0.803 (0.016) 0.835 (0.008) DCPM 0.873 (0.038) 0.906 (0.012) 0.794 (0.019) 0.982 (0.005) 0.818 (0.006) 0.828 (0.012) 0.855 (0.006) Table 2 Comparison of DCPM and fingerprint on regression task Method Regression (RMSE) ESOL Lipophilicity FreeSolv ecfp 1.342 (0.075) 0.979 (0.047) 2.149 (0.431) DCPM 0.704 (0.085) 0.877 (0.052) 1.491 (0.283) Furthermore, we constructed models for 97 ADMET property prediction tasks, significantly surpassing the limitations of contemporary methodologies, thereby facilitating a more exhaustive and nuanced prediction of molecular ADMET characteristics. Meanwhile, in comparison to traditional fingerprint-based methods, large model-based approaches outperformed the traditional models in almost 70% of the tasks. The performance of our classification models, as measured by AUC values, averaging at 0.83, while regression models demonstrated PCC values averaging 0.70 (see the Supplementary Information ADMET_cla and ADMET_reg for detailed values). We conducted a comparative analysis of endpoint information and data size across several prominent ADMET prediction platforms, namely ADMETlab 3.0[ 24 ], admetSAR 3.0[ 25 ], ProTox3.0[ 26 ], ADMETboost[ 27 ], toxCSM[ 28 ] and Interpretable-ADMET[ 29 ]. The comprehensive details of this comparison are outlined in Table 3 . The results indicate that our platform encompasses the highest number of ADMET regression tasks, aimed at achieving more accurate prediction outcomes. Furthermore, we possess the largest dataset, which serves as a robust foundation for model development. Table 3 Comparison of DCPM-ADMET with other web-based platforms name data size(compounds) endpoint classification task regression task URL DCPM-ADMET 465470 133 54 43 http://admet.bioai-global.com/ ADMETlab 3.0 400000 119 59 18 https://admetlab3.scbdd.com admetSAR3.0 370000 119 90 17 http://lmmd.ecust.edu.cn/admetsar3/ ProTox 3.0 100000 61 61 0 https://tox.charite.de/protox3/ ADMETboost 100000 22 13 9 https://ai-druglab.smu.edu/admet Interpretable-ADMET 250729 59 90 28 http://cadd.pharmacy.nankai.edu.cn/interpretableadmet/ toxCSM 43236 36 31 5 http://biosig.lab.uq.edu.au/toxcsm Implementation DCPM-ADMET presents a user-centric web application that seamlessly integrates pre-trained and 97 specialized ADMET models and 36 directly computable properties in the cloud. This platform enables users to effortlessly engage with ADMET prediction outcomes through a highly intuitive graphical interface, as exemplified in Fig. 2 . In the DCPM-ADMET prediction module, users input molecular representations in SMILES format, and the system processes them via cloud-based RDKit to generate key molecular attributes, including molecular structure, molecular weight and other properties. ADMET predictions are provided for both regression and classification tasks. Regression results include predictions for six critical aspects: basic properties, as well as absorption, distribution, metabolism, excretion, and toxicity. Drug-likeness is assessed based on thresholds derived from pharmacological research, marked as “√” (acceptable), “×” (unacceptable), or “warning” for intermediate values. For instance(Fig. 3 a), Apixaban’s bioavailability is predicted at 50.37%, marked as “√”, indicating good bioavailability, while its plasma protein binding is 86.37%, marked as “×”, suggesting reduced efficacy. In classification tasks, probabilities for various ADMET categories are displayed, with values between 0–35% labeled as negative (“-”), 65–100% as positive (“+”), and 35–65% as “warning”. For example(Fig. 3 b), Apixaban’s predictions for CYP450 2C9 inhibitor, CYP450 3A4 inhibitor, and CYP450 3A4 substrate are marked as “+”, indicating metabolism via these enzymes, consistent with its known metabolic pathways. Case Study We analyzed 45 withdrawn drugs from DrugBank[ 31 ] (version 5.1.12) and predicted their ADMET properties using DCPM-ADMET. These drugs were withdrawn due to potential toxicity, with hERG toxicity[ 32 ] and hepatotoxicity[ 33 ] being the main causes (see Supplementary Information withdrawn drugs for details). Among the 45 drugs, 24 were withdrawn due to hERG toxicity, 16 due to hepatotoxicity, 2 due to carcinogenicity and genotoxicity, and 1 due to skin irritation. Prediction of classification Properties DCPM-ADMET accurately predicted potential toxicity in 36 out of 45 drugs, marking a success rate of approximately 80.0% (see Supplementary Information withdrawn drugs for details). To ensure unbiased results, we ensured no overlap between DrugBank molecules used in these tasks and our training sets. Molecular fingerprint similarity analysis using Tanimoto coefficients revealed low average similarities (0.06–0.164) between the 45 drugs and their training set counterparts, further validating the model's generalization. Among drugs with maximum similarity below 0.4, prediction accuracy remained consistent with the overall accuracy (87.5%), reinforcing DCPM-ADMET's robust generalization capabilities (see Supplementary Information withdrawn drugs for details). Identify adverse properties Thalidomide, initially synthesized by CIBA (predecessor of Novartis), was marketed in Europe in 1957 as an antiemetic for pregnant women, tragically leading to numerous birth defects. Our tool foresaw its genotoxicity (Fig. 4 a), emphasizing the potential to avert such disasters with early predictions. Integrated Property Analysis For hERG inhibition and hepatotoxicity, DCPM-ADMET boasted prediction accuracies of 91.6% and 75.0%, respectively, underscoring its robust predictive capabilities. Specifically, Nefazodone has demonstrated the trustworthiness of DCPM-ADMET’s prediction. According to its description, Nefazodone has a very low incidence of liver damage, with a serious liver injury occurrence rate of 1 in 250,000 to 300,000 users annually. However, DCPM-ADMET is still capable of predicting its hepatotoxicity (Fig. 4 b). On the other hand, Nefazodone was predicted as high risk for DILI (drug-induced liver injury) with DCPM-ADMET, prompting further investigation. Reports suggest that Nefazodone is oxidized by CYP3A4 (our prediction also shows it as a substrate for CYP3A4), and its active metabolites can bind with GSH to produce toxicity, potentially explaining the high DILI risk and the discrepancy in its hepatotoxicity prediction. This demonstrates the tool’s ability to accurately predict ADMET properties and guide our understanding of the intrinsic factors underlying drug properties. Comparative Prediction for Multiple Drugs DCPM-ADMET predicted that Fexofenadine (0.826) has a lower risk of side effects compared to Terfenadine (0.869) (Fig. 5 ), aligning with its safer profile and reduced risk of QT prolongation. In the case of IAP antagonists, DCPM-ADMET accurately predicted the impact of structural modifications on the ADMET properties of tolinapant. By incorporating a hydroxymethyl group, tolinapant's lipophilicity decreased (LogD: 2.12 vs. 2.58; LogP: 4.17 vs. 4.32), and its probability of being a CYP3A4 substrate (60.56%) was lower than that of AT-IAP (67.09%). These predictions were consistent with experimental data 34 ., validating DCPM-ADMET's effectiveness in drug optimization. Prediction of Regression Properties To assess DCPM-ADMET's performance further, we analyzed 74 drugs with Caco-2 permeability data from DrugBank. Caco-2 permeability gauges a drug's intestinal absorption, serving as the gold standard for in vitro predictions[ 35 ]. Our model achieved an RMSE of 0.47 and a PCC of 0.82 on these data (Fig. 6 ), signifying minimal error and strong correlation between predictions and experimental values. Molecules in the test set exhibited low mean similarity to the training set (0.045–0.16), highlighting the model's generalization ability (all data and prediction values see Supplementary Information caco-2 for details). Conclusion In this study, we present DCPM-ADMET, an advanced tool for predicting drug ADMET properties using a novel pre-trained model architecture that combines the strengths of both RNN and Transformer models. By incorporating molecular substructure information through ECFP fingerprints, DCPM outperforms traditional molecular fingerprint methods, offering more accurate and reliable predictions across a wide range of molecular properties. DCPM-ADMET can calculate 133 endpoints, including 54 classification and 43 regression tasks, trained on a robust dataset of 465,470 entries related to ADMET properties. It also encompasses 36 computational properties relevant to physicochemical attributes, medicinal chemistry principles, and toxicophore rules. These extensive capabilities make DCPM-ADMET a significant advancement over existing methods, enabling early identification of drug candidates with unfavorable pharmacokinetic profiles and optimizing molecular structures for enhanced drug efficacy and safety. Additionally, its cloud-based web application provides easy access, empowering researchers to make swift, informed decisions during drug development. Looking forward, the methodologies employed in DCPM can be extended to other areas of drug discovery, such as predicting drug-drug interactions and optimizing molecular bioactivities. With the continuous availability of more data, the model's accuracy and generalizability can be further refined. Declarations Funding No Availability of data and materials All code are available at https://github.com/zhangzhangleilei/DCPM-ADMET Competing interests The authors declare that they have no competing interests. Acknowledgements We acknowledge the contributions of various individuals and organizations that have made this study possible. And the robotic AI-Scientist platform of the Chinese Academy of Sciences. Author contributions Yuchen Zeng, Yue Qi, Leilei Zhang and Kaili Jiang participated in project operations and article writing. Xiaofei Zho participated in the project planning process. Lu Liang was involved in project operations and guidance. Jianping Lin participated in project planning and article writing. References Chen M, Yang J, Tang C, Lu X, Wei Z, Liu Y, Yu P, Li H Improving ADMET Prediction Accuracy for Candidate Drugs: Factors to Consider in QSPR Modeling Approaches. Curr Top Med Chem 24 (3), 222–242. https://doi.org/10.2174/0115680266280005231207105900 Wong CH, Siah KW, Lo AW (2019) Estimation of Clinical Trial Success Rates and Related Parameters. Biostatistics 20(2):273–286. https://doi.org/10.1093/biostatistics/kxx069 Tran TTV, Tayara H, Chong KT (2023) Recent Studies of Artificial Intelligence on In Silico Drug Distribution Prediction. Int J Mol Sci 24(3):1815. https://doi.org/10.3390/ijms24031815 Tran TTV, Tayara H, Chong KT (2023) Recent Studies of Artificial Intelligence on In Silico Drug Absorption. J Chem Inf Model 63(20):6198–6211. https://doi.org/10.1021/acs.jcim.3c00960 Lv Q, Zhou F, Liu X, Zhi L (2023) Artificial Intelligence in Small Molecule Drug Discovery from 2018 to 2023: Does It Really Work? Bioorg Chem 141:106894. https://doi.org/10.1016/j.bioorg.2023.106894 Dulsat J, López-Nieto B, Estrada-Tejedor R, Borrell JI (2023) Evaluation of Free Online ADMET Tools for Academic or Small Biotech Environments. Molecules 28(2):776. https://doi.org/10.3390/molecules28020776 Dong J, Wang N-N, Yao Z-J, Zhang L, Cheng Y, Ouyang D, Lu A-P, Cao D-S (2018) ADMETlab: A Platform for Systematic ADMET Evaluation Based on a Comprehensively Collected ADMET Database. J Cheminform 10(1):29. https://doi.org/10.1186/s13321-018-0283-x Cheng F, Li W, Zhou Y, Shen J, Wu Z, Liu G, Lee PW, Tang Y, admetSAR (2012) A Comprehensive Source and Free Tool for Assessment of Chemical ADMET Properties. J Chem Inf Model 52(11):3099–3105. https://doi.org/10.1021/ci300367a Wei Y, Li S, Li Z, Wan Z, Lin J (2022) Interpretable-ADMET: A Web Service for ADMET Prediction and Optimization Based on Deep Neural Representation. Bioinformatics 38(10):2863–2871. https://doi.org/10.1093/bioinformatics/btac192 Zhang S, Yan Z, Huang Y, Liu L, He D, Wang W, Fang X, Zhang X, Wang F, Wu H, Wang H (2022) HelixADMET: A Robust and Endpoint Extensible ADMET System Incorporating Self-Supervised Knowledge Transfer. Bioinformatics 38(13):3444–3453. https://doi.org/10.1093/bioinformatics/btac342 Wu Z, Ramsundar B, Feinberg EN, Gomes J, Geniesse C, Pappu AS, Leswing K, Pande V (2018) MoleculeNet: A Benchmark for Molecular Machine Learning. Chem Sci 9(2):513–530. https://doi.org/10.1039/C7SC02664A Winter R, Montanari F, Noé F, Clevert D-A (2019) Learning Continuous and Data-Driven Molecular Descriptors by Translating Equivalent Chemical Representations. Chem Sci 10(6):1692–1701. https://doi.org/10.1039/C8SC04175J Lin X, Xu C, Xiong Z, Zhang X, Ni N, Ni B, Chang J, Pan R, Wang Z, Yu F, Tian Q, Jiang H, Zheng M, Qiao N (2023) PanGu Drug Model: Learn a Molecule like a Human. Sci China Life Sci 66(4):879–882. https://doi.org/10.1007/s11427-022-2239-y Kim S, Chen J, Cheng T, Gindulyte A, He J, He S, Li Q, Shoemaker BA, Thiessen PA, Yu B, Zaslavsky L, Zhang J, Bolton EE (2023) PubChem 2023 Update. Nucleic Acids Res 51(D1):D1373–D1380. https://doi.org/10.1093/nar/gkac956 Zdrazil B, Felix E, Hunter F, Manners EJ, Blackshaw J, Corbett S, de Veij M, Ioannidis H, Lopez DM, Mosquera JF, Magarinos MP, Bosc N, Arcila R, Kizilören T, Gaulton A, Bento AP, Adasme MF, Monecke P, Landrum GA, Leach AR (2024) The ChEMBL Database in 2023: A Drug Discovery Platform Spanning Multiple Bioactivity Data Types and Time Periods. Nucleic Acids Res 52(D1):D1180–D1192. https://doi.org/10.1093/nar/gkad1004 Huang K, Fu T, Gao W, Zhao Y, Roohani Y, Leskovec J, Coley CW, Xiao C, Sun J, Zitnik M (2021) Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development. Proceedings of Neural Information Processing Systems, NeurIPS Datasets and Benchmarks Ross J, Belgodere B, Chenthamarakshan V, Padhi I, Mroueh Y, Das P (2022) Large-Scale Chemical Language Representations Capture Molecular Structure and Properties. Nat Mach Intell 4(12):1256–1264. https://doi.org/10.1038/s42256-022-00580-7 Li H, Zhang R, Min Y, Ma D, Zhao D, Zeng JA (2023) Knowledge-Guided Pre-Training Framework for Improving Molecular Representation Learning. Nat Commun 14(1):7568. https://doi.org/10.1038/s41467-023-43214-1 Devlin J, Chang M-W, Lee K, Toutanova KBERT (2019) Pre-Training of Deep Bidirectional Transformers for Language Understanding. arXiv May 24. https://doi.org/10.48550/arXiv.1810.04805 Yang Z, Dai Z, Yang Y, Carbonell J, Salakhutdinov R, Le QV (2020) XLNet: Generalized Autoregressive Pretraining for Language Understanding. arXiv January 2. https://doi.org/10.48550/arXiv.1906.08237 Dai Z, Yang Z, Yang Y, Carbonell J, Le QV, Salakhutdinov R (2019) Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context. arXiv June 2, http://arxiv.org/abs/1901.02860 (accessed 2023-06-03). Rogers D, Hahn M, Extended-Connectivity, Fingerprints (2010) J Chem Inf Model 50(5):742–754. https://doi.org/10.1021/ci100050t Yang Z, Yang Z, Lu A, Hou T, Cao D (2021) Scopy: an integrated negative design python library for desirable HTS/VS database design. Brief Bioinform 22(3):1–15. https://doi.org/10.1093/bib/bbaa194 Fu L, Shi S, Yi J, Wang N, He Y, Wu Z, Peng J, Deng Y, Wang W, Wu C, Lyu A, Zeng X, Zhao W, Hou T, Cao D (2024) ADMETlab 3.0: An Updated Comprehensive Online ADMET Prediction Platform Enhanced with Broader Coverage, Improved Performance, API Functionality and Decision Support. Nucleic Acids Res 52(W1):W422–W431. https://doi.org/10.1093/nar/gkae236 Gu Y, Yu Z, Wang Y, Chen L, Lou C, Yang C, Li W, Liu G, Tang Y (2024) admetSAR3.0: A Comprehensive Platform for Exploration, Prediction and Optimization of Chemical ADMET Properties. Nucleic Acids Res 52(W1):W432–W438. https://doi.org/10.1093/nar/gkae298 Priyanka B, Emanuel K, Mathias D, Robert P (2024) ProTox 3.0: A Webserver for the Prediction of Toxicity of Chemicals. Nucleic Acids Res 52 Tian H, Ketkar R, Tao P, ADMETboost: (2022) A Web Server for Accurate ADMET Prediction. J Mol Model 28(12):408. https://doi.org/10.1007/s00894-022-05373-8 De Sá AGC, Long Y, Portelli S, Pires DEV, Ascher DB, toxCSM (2022) Comprehensive Prediction of Small Molecule Toxicity Profiles. Brief Bioinform 23(5):bbac337. https://doi.org/10.1093/bib/bbac337 Wei Y, Li S, Li Z, Wan Z, Lin J (2022) Interpretable-ADMET: A Web Service for ADMET Prediction and Optimization Based on Deep Neural Representation. Bioinformatics 38(10):2863–2871. https://doi.org/10.1093/bioinformatics/btac192 Chadha A, Guirguis M, Bungard TJ (2022) DOAC Drug Interactions Management Resource. Can Pharm J (Ott) 155(6):315–325. https://doi.org/10.1177/17151635221116100 Knox C, Wilson M, Klinger CM, Franklin M, Oler E, Wilson A, Pon A, Cox J, Chin NEL, Strawbridge SA, Garcia-Patino M, Kruger R, Sivakumaran A, Sanford S, Doshi R, Khetarpal N, Fatokun O, Doucet D, Zubkowski A, Rayat DY, Jackson H, Harford K, Anjum A, Zakir M, Wang F, Tian S, Lee B, Liigand J, Peters H, Wang RQR, Nguyen T, So D, Sharp M, da Silva R, Gabriel C, Scantlebury J, Jasinski M, Ackerman D, Jewison T, Sajed T, Gautam V, Wishart D (2024) DrugBank 6.0: The DrugBank Knowledgebase for 2024. Nucleic Acids Res 52(D1):D1265–D1275. https://doi.org/10.1093/nar/gkad976 Nogawa H, Kawai T (2014) hERG Trafficking Inhibition in Drug-Induced Lethal Cardiac Arrhythmia. Eur J Pharmacol 741:336–339. https://doi.org/10.1016/j.ejphar.2014.06.044 Njoku DB, Drug-Induced, Hepatotoxicity (2014) Metabolic, Genetic and Immunological Basis. Int J Mol Sci 15(4):6990–7003. https://doi.org/10.3390/ijms15046990 Martins V, Ward G, Smyth T, Reader M, Chessari G, Johnson C, Wilsher N (2024) The preclinical pharmacokinetics of Tolinapant-A dual cIAP1/XIAP antagonist with in vivo efficacy. Pharmacol Res Perspect 12:e70030. https://doi.org/10.1002/prp2.70030 Hubatsch I, Ragnarsson EGE, Artursson P (2007) Determination of Drug Permeability and Prediction of Drug Absorption in Caco-2 Monolayers. Nat Protoc 2(9):2111–2119. https://doi.org/10.1038/nprot.2007.303 Additional Declarations No competing interests reported. Supplementary Files SupplementaryInformation.xlsx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 30 Mar, 2026 Reviews received at journal 25 Mar, 2026 Reviews received at journal 19 Mar, 2026 Reviewers agreed at journal 16 Mar, 2026 Reviewers agreed at journal 24 Feb, 2026 Reviewers invited by journal 17 Dec, 2025 Editor assigned by journal 11 Dec, 2025 Submission checks completed at journal 11 Dec, 2025 First submitted to journal 10 Dec, 2025 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-8326331","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":562380238,"identity":"90dbc068-7e3e-462c-bac7-22dff2f018d2","order_by":0,"name":"Yuchen Zeng","email":"","orcid":"","institution":"Nankai University","correspondingAuthor":false,"prefix":"","firstName":"Yuchen","middleName":"","lastName":"Zeng","suffix":""},{"id":562380240,"identity":"55e889ee-f27b-4fe6-9cf0-871f05df7c07","order_by":1,"name":"Yue Qi","email":"","orcid":"","institution":"Nankai University","correspondingAuthor":false,"prefix":"","firstName":"Yue","middleName":"","lastName":"Qi","suffix":""},{"id":562380241,"identity":"a772e5f7-f1ef-4c30-9fcf-225276d9fd2e","order_by":2,"name":"Leilei Zhang","email":"","orcid":"","institution":"Nankai University","correspondingAuthor":false,"prefix":"","firstName":"Leilei","middleName":"","lastName":"Zhang","suffix":""},{"id":562380243,"identity":"c64ba543-d9bd-4093-8fd6-fc5fcc21623a","order_by":3,"name":"Kaili Jiang","email":"","orcid":"","institution":"Nankai University","correspondingAuthor":false,"prefix":"","firstName":"Kaili","middleName":"","lastName":"Jiang","suffix":""},{"id":562380245,"identity":"278a40e2-fe88-47e8-84ab-44e13e145686","order_by":4,"name":"Xiaofei Zhou","email":"","orcid":"","institution":"Tianjin BioAI-global Technology CO.,Ltd","correspondingAuthor":false,"prefix":"","firstName":"Xiaofei","middleName":"","lastName":"Zhou","suffix":""},{"id":562380247,"identity":"599bafc2-d046-4dd8-a26b-c6f79fd6e093","order_by":5,"name":"Lu Liang","email":"","orcid":"","institution":"Tianjin BioAI-global Technology CO.,Ltd","correspondingAuthor":false,"prefix":"","firstName":"Lu","middleName":"","lastName":"Liang","suffix":""},{"id":562380249,"identity":"78f270bd-22bc-4f2f-bf61-4a76d35472f4","order_by":6,"name":"Jianping Lin","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCklEQVRIiWNgGAWjYBACxmYQacDAwMfMwPgAzGQGkjzEaGFjZmA2IEoLHLABkQRENwEtzO3Mzx4XFNyxa2PnPVZdUHPHmL8d6MK3bQzy5jgdxmZuPMPgWXIbM1/a7RnHnplJHGZgNpzbxmC4swGnX8ykeQwOJ7Mx85jd5mE7bMNwmIFNmreNIcHgAC4t7N/gWop5/h22kT/MwP4bvxYesC12IC3MvG2HzQyAtjAT0FImPcPgcAJQi7E0b99hY8PDjM2Sc85JGG7AocWw//g26YI/h+35+c8Yfub5dthw3vnDBz+8KbORx2WLITBYQHGXiBQ8jCC2BHb1QCDPANFij1PFKBgFo2AUjAIAo6ZNAfSS6jgAAAAASUVORK5CYII=","orcid":"","institution":"Nankai University","correspondingAuthor":true,"prefix":"","firstName":"Jianping","middleName":"","lastName":"Lin","suffix":""}],"badges":[],"createdAt":"2025-12-10 10:38:27","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8326331/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8326331/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98642501,"identity":"32274cec-9c91-47ff-bfeb-1025a29bfe10","added_by":"auto","created_at":"2025-12-19 19:08:02","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2374190,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/3e534acddcb0171d23182cc7.docx"},{"id":98775735,"identity":"5f328204-bd24-4469-b722-851bee5a06d0","added_by":"auto","created_at":"2025-12-22 12:20:56","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7876,"visible":true,"origin":"","legend":"","description":"","filename":"2ab516475260431f93156b30df3fe294.json","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/f60b88a1fda2b6e4bb551aca.json"},{"id":98775724,"identity":"3abd8f09-9ed4-42ec-9318-c2a66a0fec82","added_by":"auto","created_at":"2025-12-22 12:20:55","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":40148,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformation.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/edb5fdf57f958e38205d06ea.xlsx"},{"id":98642494,"identity":"3b2efe26-38c3-40d8-9b1b-cbc237a214db","added_by":"auto","created_at":"2025-12-19 19:08:02","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":117313,"visible":true,"origin":"","legend":"","description":"","filename":"2ab516475260431f93156b30df3fe2941enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/a2bacd35784296885822da40.xml"},{"id":98775698,"identity":"6eb20da1-4759-4571-a213-901a0f7e8ba0","added_by":"auto","created_at":"2025-12-22 12:20:51","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":532462,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/89e317f77e7429e1104f3306.jpeg"},{"id":98775091,"identity":"3c45e326-2e78-4514-920f-19a766d69e31","added_by":"auto","created_at":"2025-12-22 12:18:24","extension":"jpeg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2550578,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/a796c77f72b25541e3a00b9a.jpeg"},{"id":98775461,"identity":"832c9236-6b04-4dc9-8725-fe2529f32286","added_by":"auto","created_at":"2025-12-22 12:20:03","extension":"jpeg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139862,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/e8c9d7c5b3d797d2617e11dd.jpeg"},{"id":98775209,"identity":"4712fe13-72b8-4056-8e7b-5ea4352651b8","added_by":"auto","created_at":"2025-12-22 12:18:54","extension":"jpeg","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":486748,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/3b189eaac3fbf2e2f863164c.jpeg"},{"id":98642504,"identity":"eb11cc55-eec1-470e-870f-e3fca24d2dd8","added_by":"auto","created_at":"2025-12-19 19:08:02","extension":"jpeg","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":931666,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/7e7717d69653ae96603a0a66.jpeg"},{"id":98775509,"identity":"c3e573b3-afd4-4123-a0c2-3507ffb20505","added_by":"auto","created_at":"2025-12-22 12:20:09","extension":"jpeg","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":232712,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/3143282ac9b0dc123f669658.jpeg"},{"id":98775236,"identity":"a0c231bf-0995-4dfe-85fe-439e038af227","added_by":"auto","created_at":"2025-12-22 12:18:59","extension":"xml","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":115173,"visible":true,"origin":"","legend":"","description":"","filename":"2ab516475260431f93156b30df3fe2941structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/38e0ebb2fefedfa6c6e557b8.xml"},{"id":98642498,"identity":"b53004b1-fc9e-442b-99d4-273a68aacf06","added_by":"auto","created_at":"2025-12-19 19:08:02","extension":"html","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":126878,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/073f3f19dc20ccaea73433e0.html"},{"id":98775277,"identity":"3160b5d3-0066-4c23-bbbe-7ffed00db87f","added_by":"auto","created_at":"2025-12-22 12:19:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":60961,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Pre-training process of DCPM. (b) Fine-tuning and inference flow of DCPM on property data set\u003c/p\u003e","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/67ee42d1aafb8147ac108c31.png"},{"id":98642486,"identity":"f0aae682-57c4-43f7-bcd2-b26b8a26d1ec","added_by":"auto","created_at":"2025-12-19 19:08:01","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":29990,"visible":true,"origin":"","legend":"\u003cp\u003eDCPM-ADMET Web Application Prediction Workflow\u003c/p\u003e","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/75b4c1ca7bbe917f3aa15c1c.png"},{"id":98775721,"identity":"f95f2e7f-f75a-4fa5-a7fa-4a510e29cfa9","added_by":"auto","created_at":"2025-12-22 12:20:55","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":20243,"visible":true,"origin":"","legend":"\u003cp\u003eApixaban Prediction Results\u003c/p\u003e","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/532d6ec7d8d00395adbb0f1d.png"},{"id":98642493,"identity":"d4ff2d4b-fc29-46f4-a8a0-83e7b1c439d2","added_by":"auto","created_at":"2025-12-19 19:08:01","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":58618,"visible":true,"origin":"","legend":"\u003cp\u003eThalidomide \u0026amp; Nefazodone Prediction Insights\u003c/p\u003e","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/b463aebf1b375014e707480c.png"},{"id":98642496,"identity":"4f3a7df5-3b4c-44c3-ae96-28d7e1c288e3","added_by":"auto","created_at":"2025-12-19 19:08:02","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":102975,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction results of a series of drugs\u003c/p\u003e","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/79acfc5765acd0b89e9c975f.png"},{"id":98775575,"identity":"87e73d4d-1768-4fcb-9e3f-ca3390c591e6","added_by":"auto","created_at":"2025-12-22 12:20:21","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":31890,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction results for 74 drugs with Caco-2 values\u003c/p\u003e","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/c9e2f303d96fb1cddf6e90ff.png"},{"id":98782621,"identity":"9147521f-65cf-4063-a512-7525ef659d73","added_by":"auto","created_at":"2025-12-22 12:40:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1194887,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/d1e7edb7-4480-4a6f-add5-13a2122b7a4a.pdf"},{"id":98642490,"identity":"d7771c2d-b0fe-421b-8d82-7daccb5a7b31","added_by":"auto","created_at":"2025-12-19 19:08:01","extension":"xlsx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":40148,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformation.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8326331/v1/c96f8acc45b08d1a84ca2065.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"DCPM-ADMET: Fusion of Dual-channel Pre-trained Model and Molecular Fingerprints to enhance Drug ADMET Properties Prediction","fulltext":[{"header":"Introduction","content":"\u003cp\u003eDrug development is a complex, time-consuming, and expensive process. Traditional drug development methods often require years and substantial financial investment to bring a new drug to market[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The pivotal success of clinical trials significantly shapes the course of drug development, holding paramount importance for pharmaceutical investors and patients alike. Nevertheless, about 40\u0026ndash;60% of drug candidates succumb to unfavorable ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) characteristics during clinical trials[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. By anticipating these ADMET properties, researchers can pinpoint and eradicate unfavorable pharmacokinetic traits in prospective compounds, expediting the developmental process while curbing time and financial expenses. This proactive strategy also facilitates the early recognition of potentially efficacious compounds during screening, prioritizing them for subsequent advancements and refining the selection of candidates with the highest likelihood of success as novel medications. Furthermore, it underpins the determination of optimal dosing, optimal routes of administration, and minimizes the threat of adverse reactions, enhancing both the safety and efficacy of drug candidates[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e][\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn recent years, the landscape of computational simulation software for ADMET property prediction has undergone remarkable advancements, achieving notable strides in enhancing prediction accuracy[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. To further refine this intricate endeavor, artificial intelligence technologies have emerged as pivotal players in diverse ADMET computational frameworks, yielding an array of outstanding prediction tools, including ADMETlab, admetSAR, Interpretable-ADMET, HelixADMET, among others[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e][\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e][\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e][\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e][\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Nevertheless, the application of current models confronts myriad challenges, encompassing poor data quality, limited generalization capabilities, interpretability gaps, data scarcity and uncertainties, thereby restricting model accuracy and reliability. These hurdles underscore the urgency for continued research and optimization.\u003c/p\u003e \u003cp\u003eThe algorithmic evolution in drug ADMET prediction has witnessed profound transitions, evolving from traditional machine learning through deep learning to the advent of pre-trained models. Initially, researchers harnessed traditional methods like linear regression, support vector machines, and random forests, relying on manual feature engineering to capture molecular attributes[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. However, these approaches often grappled with the intricacies of ADMET prediction. As computational prowess and data availability soared, deep learning techniques emerged, automating feature learning and demonstrating exceptional performance in ADMET prediction tasks[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Nevertheless, their hunger for massive datasets and challenges in generalization amidst data scarcity remain obstacles. Recently, pre-trained models have emerged as promising avenues for ADMET prediction. By pre-training on extensive datasets, these models acquire rich feature representations that are subsequently fine-tuned for specific ADMET tasks, significantly bolstering generalization and prediction precision. They not only adapt seamlessly to diverse ADMET prediction scenarios but also address data scarcity and uncertainty more adeptly, marking the cutting edge of algorithmic developments in this field[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn the present work, the model underwent an extensive unsupervised pre-training regimen utilizing a vast small molecule dataset. The pre-training employs a dual-channel architecture to discern intricate and abstract feature representations from different aspects, which we refer to as Dual-channel Pre-trained Model(DCPM). One channel, based on the XLNet architecture, is used to learn the semantic differences in molecular sequences, while the other, utilizing an RNN-based architecture, captures complex representations at the level of molecular properties. Subsequently, DCPM was trained and evaluated across a broad spectrum of downstream tasks, encompassing various challenges within the MoleculeNet benchmark dataset, such as predicting Caco-2 Permeability, BBB Penetration, and identifying CYP450 1A2 Inhibitors. Its state-of-the-art performance and robust generalization capabilities across these tasks validated its accuracy and reliability. Building upon this success, we further extended DCPM's reach by subjecting it to rigorous training and testing within 97 downstream ADMET tasks. We conducted a comparison with the traditional molecular fingerprint ECFP and our method demonstrated superiority in nearly 70% of the tasks. In addition, when compared with existing tools, we encompass the largest number of ADMET regression tasks designed to deliver more precise prediction outcomes, and simultaneously, our overall dataset is also the largest, containing over 460 000 entries, thereby laying a solid foundation for a more holistic and comprehensive approach to drug ADMET prediction.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003eData Preparation\u003c/p\u003e\n\u003cdiv id=\"Sec3\"\u003e\n \u003ch2\u003ePre-training data\u003c/h2\u003e\n \u003cp\u003eThe DCPM-ADMET training process incorporates diverse datasets, encompassing chemical molecule collections for pre-training and targeted ADMET datasets for constructing predictive models. The pre-training phase leverages PubChem[14], a comprehensive bioactivity database housing millions of compound records spanning pharmaceuticals, natural products, and chemicals. From this repository, we extracted SMILES and InChI representations of chemical molecules, subjecting them to a rigorous data preparation pipeline:\u003c/p\u003e\n \u003cp\u003e1. Data Standardization: Utilizing RDKit (version 2023.9.6), SMILES strings underwent standardization procedures, including bond order canonicalization and removal of redundant hydrogen atoms.\u003c/p\u003e\n \u003cp\u003e2. Missing \u0026amp; Duplicate Value Filtering: Molecules that failed standardization were deemed missing values, the missing and duplicate values are removed to ensure data uniqueness and validity.\u003c/p\u003e\n \u003cp\u003e3. Descriptor Calculation: Based on the standardized SMILES, we employed RDKit to compute nine crucial physicochemical properties for each molecule, serving as molecular descriptors. These encompassed logP, maximal and minimal partial charges, valence electron count, hydrogen bond donor and acceptor numbers, Balaban\u0026apos;s J value, molar refractivity, and topological polar surface area.\u003c/p\u003e\n \u003cp\u003eThese processes yielded 111 295 885 data points for pre-training.\u003c/p\u003e\n\u003c/div\u003e\n\u003ch3\u003eADMET datasets\u003c/h3\u003e\n\u003cp\u003eFor the ADMET property datasets, we aggregated resources from multiple experimental compound databases, notably PubChem, ChEMBL[15] and TDC[16], and implemented a meticulous data processing strategy:\u003c/p\u003e\n\u003cp\u003e1. Data Standardization: SMILES representations were standardized using RDKit, and harmonizing data across datasets into consistent units like pEC50 and mol/L.\u003c/p\u003e\n\u003cp\u003e2. Data Cleaning \u0026amp; Merging: Rigorous comparison of entries across similar datasets from different sources led to the removal of conflicting samples in classification sets and unreliable samples in regression sets exhibiting discrepancies exceeding an order of magnitude. Subsequently, datasets were merged and further cleaned.\u003c/p\u003e\n\u003cp\u003e3. Duplicate Data Elimination: Following cleaning and merging, duplicate entries were purged, resulting in a refined ADMET dataset.\u003c/p\u003e\n\u003cp\u003eThis final dataset comprises 97 ADMET sets, segmented into 43 regression and 54 classification datasets, totaling 465 470 data entries.\u003c/p\u003e\n\u003ch3\u003eModel Architecture\u003c/h3\u003e\n\u003cp\u003eThe DCPM-ADMET molecular property prediction framework encapsulates two essential components: a feature extraction module and a downstream task prediction module. The feature extraction module meticulously crafts deep and versatile vector representations from SMILES molecular sequences. These representations then underpin tailored modeling for diverse downstream tasks.\u003c/p\u003e\n\u003ch3\u003ePre-training\u003c/h3\u003e\n\u003cp\u003eInspired by Transformer-based models\u0026apos; exceptional performance, DCPM embraces pre-trained architectures to refine SMILES sequences. However, prevalent methods like MoLFormer[17] and KPGT[18], which adopt the Transformer encoder during pre-training with masked token recovery akin to BERT[19], confront a pre-training-fine-tuning disparity when real-world inputs lack mask tokens[20].\u003c/p\u003e\n\u003cp\u003e1. XLNet[21], innovatively combining autoregressive and permutation strategies, resolves this inconsistency while enabling bidirectional context analysis, a feat previously unattainable by autoregressive models. By leveraging XLNet\u0026apos;s permutation technique and dual-stream attention, DCPM captures global SMILES sequence patterns through self-attention and distills chemical insights from bidirectionally permuted contexts, enriching feature representation.\u003c/p\u003e\n\u003cp\u003e2. In the realm of Natural Language Processing (NLP), XLNet leverages the Transformer-XL architecture to enhance its capability in managing extremely long texts. Nevertheless, the SMILES sequences of chemical molecules, characterized by their relatively shorter lengths and smaller vocabularies, are more susceptible to overfitting challenges. To mitigate this, we have omitted the long text processing feature and drawing incorporated an additional RNN structure inspiration from the approach of Robin Winter[12]. This RNN conducts a Seq2Seq SMILES-to-InChI translation while regressing nine physicochemical descriptors. By integrating InChI\u0026apos;s fundamental physicochemical insights and descriptor regression into the pre-training objectives, we augment task complexity, mitigating overfitting, and fostering deeper molecular knowledge extraction beyond mere sequence reconstruction.\u003c/p\u003e\n\u003cp\u003e3. Considering that tokenized SMILES sequences mainly consist of atomic tokens, lacking substructure fragment information of chemical molecules, we added a feature fusion layer to DCPM. This layer amalgamates XLNet and RNN hidden layers with Extended Connectivity Fingerprints (ECFP)[22], an atomic connectivity graph-based descriptor rich in topological and substructure information, augmenting the model\u0026apos;s substructural comprehension.\u003c/p\u003e\n\u003cp\u003eIn Fig.\u0026nbsp;1a, the pre-training workflow is outlined: SMILES are tokenized, embedded, and processed by an Encoder amalgamating GRU cells and XLNet layers. Extracted features are fed into three decoders, optimizing across InChI reconstruction, property prediction, and masked structure tasks.\u003c/p\u003e\n\u003ch3\u003eFine-tuning\u003c/h3\u003e\n\u003cp\u003eFor fine-tuning, depicted in Fig.\u0026nbsp;1b, the feature extraction layer is frozen, while the downstream prediction layer is trained. For binary classification and regression tasks, we employ Random Forest to predict properties. The deep neural network applied in multi-task uses three hidden layers, optimizing key hyperparameters such as L2 regularization, dropout, and the number of neurons in each hidden layer. The Tree-structured Parzen Estimator(TPE) algorithm is used to identify the best hyperparameters across 50 evaluations. Early stopping and a random 8:1:1 train-validate-test split with AUC metrics over five runs ensure robustness.\u003c/p\u003e\n\u003cp\u003eDCPM\u0026apos;s prowess is validated against ECFP on MoleculeNet, demonstrating superiority across 10 subsets (7 classification and 3 regression tasks). Among the seven classification datasets, ClinTox, SIDER, Tox21, and ToxCast are multi-label classification task, while the other three classification datasets (BACE, BBBP, and HIV) and the regression datasets (ESOL, Lipophilicity, FreeSolv) are single-task. Ultimately, DCPM-ADMET is trained on a comprehensive ADMET dataset, encompassing 54 classification and 43 regression tasks, solidifying its standing as a potent tool for ADMET property prediction.\u003c/p\u003e\n\u003cdiv id=\"Sec8\"\u003e\n \u003ch2\u003eDirectly computable properties\u003c/h2\u003e\n \u003cp\u003eFor enhanced usability, we have integrated 36 additional directly computable endpoints (see the Supplementary Information calculated_props for detailed values), encompassing the calculation of physicochemical properties, medicinal chemistry principles, and toxicophore rules using the RDKit and Scopy packages[23].\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Results and Discussion","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eResults and Implementation of DCPM-ADMET\u003c/h2\u003e \u003cp\u003eFirst, when compared with traditional fingerprint features, ECFP, across 10 benchmark datasets, DCPM outperformed traditional fingerprints in 7 out of 10 tasks, demonstrating a clear advantage (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\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\u003eComparison of DCPM and fingerprint on classification task\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMethod\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"7\" nameend=\"c8\" namest=\"c2\"\u003e \u003cp\u003eClassification (ROC-AUC)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBACE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBBBP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHIV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eClinTox\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSIDER\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eTox21\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eToxCast\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eecfp\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.888\u003csub\u003e(0.023)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.909\u003csub\u003e(0.011)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.811\u003csub\u003e(0.015)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.978\u003csub\u003e(0.009)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.815\u003csub\u003e(0.006)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.803\u003csub\u003e(0.016)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.835\u003csub\u003e(0.008)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDCPM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.873\u003csub\u003e(0.038)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.906\u003csub\u003e(0.012)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.794\u003csub\u003e(0.019)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.982\u003c/b\u003e\u003csub\u003e\u003cb\u003e(0.005)\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.818\u003c/b\u003e\u003csub\u003e\u003cb\u003e(0.006)\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e0.828\u003c/b\u003e\u003csub\u003e\u003cb\u003e(0.012)\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e0.855\u003c/b\u003e\u003csub\u003e\u003cb\u003e(0.006)\u003c/b\u003e\u003c/sub\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\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\u003eComparison of DCPM and fingerprint on regression task\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\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 \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMethod\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eRegression (RMSE)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eESOL\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLipophilicity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFreeSolv\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eecfp\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.342\u003csub\u003e(0.075)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.979\u003csub\u003e(0.047)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.149\u003csub\u003e(0.431)\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDCPM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e0.704\u003c/b\u003e\u003csub\u003e\u003cb\u003e(0.085)\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.877\u003c/b\u003e\u003csub\u003e\u003cb\u003e(0.052)\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e1.491\u003c/b\u003e\u003csub\u003e\u003cb\u003e(0.283)\u003c/b\u003e\u003c/sub\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\u003eFurthermore, we constructed models for 97 ADMET property prediction tasks, significantly surpassing the limitations of contemporary methodologies, thereby facilitating a more exhaustive and nuanced prediction of molecular ADMET characteristics. Meanwhile, in comparison to traditional fingerprint-based methods, large model-based approaches outperformed the traditional models in almost 70% of the tasks. The performance of our classification models, as measured by AUC values, averaging at 0.83, while regression models demonstrated PCC values averaging 0.70 (see the Supplementary Information ADMET_cla and ADMET_reg for detailed values).\u003c/p\u003e \u003cp\u003eWe conducted a comparative analysis of endpoint information and data size across several prominent ADMET prediction platforms, namely ADMETlab 3.0[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], admetSAR 3.0[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], ProTox3.0[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], ADMETboost[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], toxCSM[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] and Interpretable-ADMET[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. The comprehensive details of this comparison are outlined in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The results indicate that our platform encompasses the highest number of ADMET regression tasks, aimed at achieving more accurate prediction outcomes. Furthermore, we possess the largest dataset, which serves as a robust foundation for model development.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of DCPM-ADMET with other web-based platforms\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\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=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ename\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003edata size(compounds)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eendpoint\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eclassification task\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eregression task\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eURL\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDCPM-ADMET\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e465470\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e133\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e43\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://admet.bioai-global.com/\u003c/span\u003e\u003cspan address=\"http://admet.bioai-global.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eADMETlab 3.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e400000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e119\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://admetlab3.scbdd.com\u003c/span\u003e\u003cspan address=\"https://admetlab3.scbdd.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eadmetSAR3.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e370000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e119\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://lmmd.ecust.edu.cn/admetsar3/\u003c/span\u003e\u003cspan address=\"http://lmmd.ecust.edu.cn/admetsar3/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProTox 3.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e100000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://tox.charite.de/protox3/\u003c/span\u003e\u003cspan address=\"https://tox.charite.de/protox3/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eADMETboost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e100000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ai-druglab.smu.edu/admet\u003c/span\u003e\u003cspan address=\"https://ai-druglab.smu.edu/admet\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInterpretable-ADMET\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e250729\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://cadd.pharmacy.nankai.edu.cn/interpretableadmet/\u003c/span\u003e\u003cspan address=\"http://cadd.pharmacy.nankai.edu.cn/interpretableadmet/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003etoxCSM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e43236\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://biosig.lab.uq.edu.au/toxcsm\u003c/span\u003e\u003cspan address=\"http://biosig.lab.uq.edu.au/toxcsm\" targettype=\"URL\" class=\"RefTarget\"\u003e\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 \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eImplementation\u003c/h2\u003e \u003cp\u003eDCPM-ADMET presents a user-centric web application that seamlessly integrates pre-trained and 97 specialized ADMET models and 36 directly computable properties in the cloud. This platform enables users to effortlessly engage with ADMET prediction outcomes through a highly intuitive graphical interface, as exemplified in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn the DCPM-ADMET prediction module, users input molecular representations in SMILES format, and the system processes them via cloud-based RDKit to generate key molecular attributes, including molecular structure, molecular weight and other properties. ADMET predictions are provided for both regression and classification tasks.\u003c/p\u003e \u003cp\u003eRegression results include predictions for six critical aspects: basic properties, as well as absorption, distribution, metabolism, excretion, and toxicity. Drug-likeness is assessed based on thresholds derived from pharmacological research, marked as \u0026ldquo;\u0026radic;\u0026rdquo; (acceptable), \u0026ldquo;\u0026times;\u0026rdquo; (unacceptable), or \u0026ldquo;warning\u0026rdquo; for intermediate values. For instance(Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea), Apixaban\u0026rsquo;s bioavailability is predicted at 50.37%, marked as \u0026ldquo;\u0026radic;\u0026rdquo;, indicating good bioavailability, while its plasma protein binding is 86.37%, marked as \u0026ldquo;\u0026times;\u0026rdquo;, suggesting reduced efficacy.\u003c/p\u003e \u003cp\u003eIn classification tasks, probabilities for various ADMET categories are displayed, with values between 0\u0026ndash;35% labeled as negative (\u0026ldquo;-\u0026rdquo;), 65\u0026ndash;100% as positive (\u0026ldquo;+\u0026rdquo;), and 35\u0026ndash;65% as \u0026ldquo;warning\u0026rdquo;. For example(Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb), Apixaban\u0026rsquo;s predictions for CYP450 2C9 inhibitor, CYP450 3A4 inhibitor, and CYP450 3A4 substrate are marked as \u0026ldquo;+\u0026rdquo;, indicating metabolism via these enzymes, consistent with its known metabolic pathways.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eCase Study\u003c/h2\u003e \u003cp\u003eWe analyzed 45 withdrawn drugs from DrugBank[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] (version 5.1.12) and predicted their ADMET properties using DCPM-ADMET. These drugs were withdrawn due to potential toxicity, with hERG toxicity[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] and hepatotoxicity[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] being the main causes (see Supplementary Information withdrawn drugs for details). Among the 45 drugs, 24 were withdrawn due to hERG toxicity, 16 due to hepatotoxicity, 2 due to carcinogenicity and genotoxicity, and 1 due to skin irritation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003ePrediction of classification Properties\u003c/h2\u003e \u003cp\u003eDCPM-ADMET accurately predicted potential toxicity in 36 out of 45 drugs, marking a success rate of approximately 80.0% (see Supplementary Information withdrawn drugs for details). To ensure unbiased results, we ensured no overlap between DrugBank molecules used in these tasks and our training sets. Molecular fingerprint similarity analysis using Tanimoto coefficients revealed low average similarities (0.06\u0026ndash;0.164) between the 45 drugs and their training set counterparts, further validating the model's generalization. Among drugs with maximum similarity below 0.4, prediction accuracy remained consistent with the overall accuracy (87.5%), reinforcing DCPM-ADMET's robust generalization capabilities (see Supplementary Information withdrawn drugs for details).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eIdentify adverse properties\u003c/h2\u003e \u003cp\u003eThalidomide, initially synthesized by CIBA (predecessor of Novartis), was marketed in Europe in 1957 as an antiemetic for pregnant women, tragically leading to numerous birth defects. Our tool foresaw its genotoxicity (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea), emphasizing the potential to avert such disasters with early predictions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eIntegrated Property Analysis\u003c/h2\u003e \u003cp\u003eFor hERG inhibition and hepatotoxicity, DCPM-ADMET boasted prediction accuracies of 91.6% and 75.0%, respectively, underscoring its robust predictive capabilities. Specifically, Nefazodone has demonstrated the trustworthiness of DCPM-ADMET\u0026rsquo;s prediction. According to its description, Nefazodone has a very low incidence of liver damage, with a serious liver injury occurrence rate of 1 in 250,000 to 300,000 users annually. However, DCPM-ADMET is still capable of predicting its hepatotoxicity (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb). On the other hand, Nefazodone was predicted as high risk for DILI (drug-induced liver injury) with DCPM-ADMET, prompting further investigation. Reports suggest that Nefazodone is oxidized by CYP3A4 (our prediction also shows it as a substrate for CYP3A4), and its active metabolites can bind with GSH to produce toxicity, potentially explaining the high DILI risk and the discrepancy in its hepatotoxicity prediction. This demonstrates the tool\u0026rsquo;s ability to accurately predict ADMET properties and guide our understanding of the intrinsic factors underlying drug properties.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eComparative Prediction for Multiple Drugs\u003c/h2\u003e \u003cp\u003eDCPM-ADMET predicted that Fexofenadine (0.826) has a lower risk of side effects compared to Terfenadine (0.869) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e), aligning with its safer profile and reduced risk of QT prolongation. In the case of IAP antagonists, DCPM-ADMET accurately predicted the impact of structural modifications on the ADMET properties of tolinapant. By incorporating a hydroxymethyl group, tolinapant's lipophilicity decreased (LogD: 2.12 vs. 2.58; LogP: 4.17 vs. 4.32), and its probability of being a CYP3A4 substrate (60.56%) was lower than that of AT-IAP (67.09%). These predictions were consistent with experimental data\u003csup\u003e34\u003c/sup\u003e., validating DCPM-ADMET's effectiveness in drug optimization.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003ePrediction of Regression Properties\u003c/h2\u003e \u003cp\u003eTo assess DCPM-ADMET's performance further, we analyzed 74 drugs with Caco-2 permeability data from DrugBank. Caco-2 permeability gauges a drug's intestinal absorption, serving as the gold standard for in vitro predictions[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. Our model achieved an RMSE of 0.47 and a PCC of 0.82 on these data (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e), signifying minimal error and strong correlation between predictions and experimental values. Molecules in the test set exhibited low mean similarity to the training set (0.045\u0026ndash;0.16), highlighting the model's generalization ability (all data and prediction values see Supplementary Information caco-2 for details).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this study, we present DCPM-ADMET, an advanced tool for predicting drug ADMET properties using a novel pre-trained model architecture that combines the strengths of both RNN and Transformer models. By incorporating molecular substructure information through ECFP fingerprints, DCPM outperforms traditional molecular fingerprint methods, offering more accurate and reliable predictions across a wide range of molecular properties.\u003c/p\u003e \u003cp\u003eDCPM-ADMET can calculate 133 endpoints, including 54 classification and 43 regression tasks, trained on a robust dataset of 465,470 entries related to ADMET properties. It also encompasses 36 computational properties relevant to physicochemical attributes, medicinal chemistry principles, and toxicophore rules. These extensive capabilities make DCPM-ADMET a significant advancement over existing methods, enabling early identification of drug candidates with unfavorable pharmacokinetic profiles and optimizing molecular structures for enhanced drug efficacy and safety. Additionally, its cloud-based web application provides easy access, empowering researchers to make swift, informed decisions during drug development.\u003c/p\u003e \u003cp\u003eLooking forward, the methodologies employed in DCPM can be extended to other areas of drug discovery, such as predicting drug-drug interactions and optimizing molecular bioactivities. With the continuous availability of more data, the model's accuracy and generalizability can be further refined.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll code are available at https://github.com/zhangzhangleilei/DCPM-ADMET\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe acknowledge the contributions of various individuals and organizations that have made this study possible. And the robotic AI-Scientist platform of the Chinese Academy of Sciences.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eYuchen Zeng, Yue Qi, Leilei Zhang and Kaili Jiang participated in project operations and article writing. Xiaofei Zho participated in the project planning process. Lu Liang was involved in project operations and guidance. Jianping Lin participated in project planning and article writing.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eChen M, Yang J, Tang C, Lu X, Wei Z, Liu Y, Yu P, Li H Improving ADMET Prediction Accuracy for Candidate Drugs: Factors to Consider in QSPR Modeling Approaches. Curr Top Med Chem \u003cem\u003e24\u003c/em\u003e (3), 222\u0026ndash;242. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2174/0115680266280005231207105900\u003c/span\u003e\u003cspan address=\"10.2174/0115680266280005231207105900\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWong CH, Siah KW, Lo AW (2019) Estimation of Clinical Trial Success Rates and Related Parameters. Biostatistics 20(2):273\u0026ndash;286. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/biostatistics/kxx069\u003c/span\u003e\u003cspan address=\"10.1093/biostatistics/kxx069\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTran TTV, Tayara H, Chong KT (2023) Recent Studies of Artificial Intelligence on In Silico Drug Distribution Prediction. Int J Mol Sci 24(3):1815. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ijms24031815\u003c/span\u003e\u003cspan address=\"10.3390/ijms24031815\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTran TTV, Tayara H, Chong KT (2023) Recent Studies of Artificial Intelligence on In Silico Drug Absorption. J Chem Inf Model 63(20):6198\u0026ndash;6211. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1021/acs.jcim.3c00960\u003c/span\u003e\u003cspan address=\"10.1021/acs.jcim.3c00960\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLv Q, Zhou F, Liu X, Zhi L (2023) Artificial Intelligence in Small Molecule Drug Discovery from 2018 to 2023: Does It Really Work? Bioorg Chem 141:106894. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.bioorg.2023.106894\u003c/span\u003e\u003cspan address=\"10.1016/j.bioorg.2023.106894\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDulsat J, L\u0026oacute;pez-Nieto B, Estrada-Tejedor R, Borrell JI (2023) Evaluation of Free Online ADMET Tools for Academic or Small Biotech Environments. Molecules 28(2):776. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/molecules28020776\u003c/span\u003e\u003cspan address=\"10.3390/molecules28020776\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDong J, Wang N-N, Yao Z-J, Zhang L, Cheng Y, Ouyang D, Lu A-P, Cao D-S (2018) ADMETlab: A Platform for Systematic ADMET Evaluation Based on a Comprehensively Collected ADMET Database. J Cheminform 10(1):29. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s13321-018-0283-x\u003c/span\u003e\u003cspan address=\"10.1186/s13321-018-0283-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheng F, Li W, Zhou Y, Shen J, Wu Z, Liu G, Lee PW, Tang Y, admetSAR (2012) A Comprehensive Source and Free Tool for Assessment of Chemical ADMET Properties. J Chem Inf Model 52(11):3099\u0026ndash;3105. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1021/ci300367a\u003c/span\u003e\u003cspan address=\"10.1021/ci300367a\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei Y, Li S, Li Z, Wan Z, Lin J (2022) Interpretable-ADMET: A Web Service for ADMET Prediction and Optimization Based on Deep Neural Representation. Bioinformatics 38(10):2863\u0026ndash;2871. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/bioinformatics/btac192\u003c/span\u003e\u003cspan address=\"10.1093/bioinformatics/btac192\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang S, Yan Z, Huang Y, Liu L, He D, Wang W, Fang X, Zhang X, Wang F, Wu H, Wang H (2022) HelixADMET: A Robust and Endpoint Extensible ADMET System Incorporating Self-Supervised Knowledge Transfer. Bioinformatics 38(13):3444\u0026ndash;3453. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/bioinformatics/btac342\u003c/span\u003e\u003cspan address=\"10.1093/bioinformatics/btac342\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu Z, Ramsundar B, Feinberg EN, Gomes J, Geniesse C, Pappu AS, Leswing K, Pande V (2018) MoleculeNet: A Benchmark for Molecular Machine Learning. Chem Sci 9(2):513\u0026ndash;530. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1039/C7SC02664A\u003c/span\u003e\u003cspan address=\"10.1039/C7SC02664A\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWinter R, Montanari F, No\u0026eacute; F, Clevert D-A (2019) Learning Continuous and Data-Driven Molecular Descriptors by Translating Equivalent Chemical Representations. Chem Sci 10(6):1692\u0026ndash;1701. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1039/C8SC04175J\u003c/span\u003e\u003cspan address=\"10.1039/C8SC04175J\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin X, Xu C, Xiong Z, Zhang X, Ni N, Ni B, Chang J, Pan R, Wang Z, Yu F, Tian Q, Jiang H, Zheng M, Qiao N (2023) PanGu Drug Model: Learn a Molecule like a Human. Sci China Life Sci 66(4):879\u0026ndash;882. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11427-022-2239-y\u003c/span\u003e\u003cspan address=\"10.1007/s11427-022-2239-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim S, Chen J, Cheng T, Gindulyte A, He J, He S, Li Q, Shoemaker BA, Thiessen PA, Yu B, Zaslavsky L, Zhang J, Bolton EE (2023) PubChem 2023 Update. Nucleic Acids Res 51(D1):D1373\u0026ndash;D1380. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/nar/gkac956\u003c/span\u003e\u003cspan address=\"10.1093/nar/gkac956\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZdrazil B, Felix E, Hunter F, Manners EJ, Blackshaw J, Corbett S, de Veij M, Ioannidis H, Lopez DM, Mosquera JF, Magarinos MP, Bosc N, Arcila R, Kizil\u0026ouml;ren T, Gaulton A, Bento AP, Adasme MF, Monecke P, Landrum GA, Leach AR (2024) The ChEMBL Database in 2023: A Drug Discovery Platform Spanning Multiple Bioactivity Data Types and Time Periods. Nucleic Acids Res 52(D1):D1180\u0026ndash;D1192. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/nar/gkad1004\u003c/span\u003e\u003cspan address=\"10.1093/nar/gkad1004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang K, Fu T, Gao W, Zhao Y, Roohani Y, Leskovec J, Coley CW, Xiao C, Sun J, Zitnik M (2021) Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development. \u003cem\u003eProceedings of Neural Information Processing Systems, NeurIPS Datasets and Benchmarks\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoss J, Belgodere B, Chenthamarakshan V, Padhi I, Mroueh Y, Das P (2022) Large-Scale Chemical Language Representations Capture Molecular Structure and Properties. Nat Mach Intell 4(12):1256\u0026ndash;1264. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s42256-022-00580-7\u003c/span\u003e\u003cspan address=\"10.1038/s42256-022-00580-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi H, Zhang R, Min Y, Ma D, Zhao D, Zeng JA (2023) Knowledge-Guided Pre-Training Framework for Improving Molecular Representation Learning. Nat Commun 14(1):7568. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41467-023-43214-1\u003c/span\u003e\u003cspan address=\"10.1038/s41467-023-43214-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDevlin J, Chang M-W, Lee K, Toutanova KBERT (2019) Pre-Training of Deep Bidirectional Transformers for Language Understanding. arXiv May 24. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.1810.04805\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.1810.04805\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang Z, Dai Z, Yang Y, Carbonell J, Salakhutdinov R, Le QV (2020) XLNet: Generalized Autoregressive Pretraining for Language Understanding. arXiv January 2. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.1906.08237\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.1906.08237\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDai Z, Yang Z, Yang Y, Carbonell J, Le QV, Salakhutdinov R (2019) Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context. arXiv June 2, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/1901.02860\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/1901.02860\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (accessed 2023-06-03).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRogers D, Hahn M, Extended-Connectivity, Fingerprints (2010) J Chem Inf Model 50(5):742\u0026ndash;754. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1021/ci100050t\u003c/span\u003e\u003cspan address=\"10.1021/ci100050t\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang Z, Yang Z, Lu A, Hou T, Cao D (2021) Scopy: an integrated negative design python library for desirable HTS/VS database design. Brief Bioinform 22(3):1\u0026ndash;15. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/bib/bbaa194\u003c/span\u003e\u003cspan address=\"10.1093/bib/bbaa194\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFu L, Shi S, Yi J, Wang N, He Y, Wu Z, Peng J, Deng Y, Wang W, Wu C, Lyu A, Zeng X, Zhao W, Hou T, Cao D (2024) ADMETlab 3.0: An Updated Comprehensive Online ADMET Prediction Platform Enhanced with Broader Coverage, Improved Performance, API Functionality and Decision Support. Nucleic Acids Res 52(W1):W422\u0026ndash;W431. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/nar/gkae236\u003c/span\u003e\u003cspan address=\"10.1093/nar/gkae236\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGu Y, Yu Z, Wang Y, Chen L, Lou C, Yang C, Li W, Liu G, Tang Y (2024) admetSAR3.0: A Comprehensive Platform for Exploration, Prediction and Optimization of Chemical ADMET Properties. Nucleic Acids Res 52(W1):W432\u0026ndash;W438. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/nar/gkae298\u003c/span\u003e\u003cspan address=\"10.1093/nar/gkae298\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePriyanka B, Emanuel K, Mathias D, Robert P (2024) ProTox 3.0: A Webserver for the Prediction of Toxicity of Chemicals. Nucleic Acids Res \u003cem\u003e52\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTian H, Ketkar R, Tao P, ADMETboost: (2022) A Web Server for Accurate ADMET Prediction. J Mol Model 28(12):408. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00894-022-05373-8\u003c/span\u003e\u003cspan address=\"10.1007/s00894-022-05373-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe S\u0026aacute; AGC, Long Y, Portelli S, Pires DEV, Ascher DB, toxCSM (2022) Comprehensive Prediction of Small Molecule Toxicity Profiles. Brief Bioinform 23(5):bbac337. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/bib/bbac337\u003c/span\u003e\u003cspan address=\"10.1093/bib/bbac337\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei Y, Li S, Li Z, Wan Z, Lin J (2022) Interpretable-ADMET: A Web Service for ADMET Prediction and Optimization Based on Deep Neural Representation. Bioinformatics 38(10):2863\u0026ndash;2871. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/bioinformatics/btac192\u003c/span\u003e\u003cspan address=\"10.1093/bioinformatics/btac192\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChadha A, Guirguis M, Bungard TJ (2022) DOAC Drug Interactions Management Resource. Can Pharm J (Ott) 155(6):315\u0026ndash;325. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/17151635221116100\u003c/span\u003e\u003cspan address=\"10.1177/17151635221116100\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKnox C, Wilson M, Klinger CM, Franklin M, Oler E, Wilson A, Pon A, Cox J, Chin NEL, Strawbridge SA, Garcia-Patino M, Kruger R, Sivakumaran A, Sanford S, Doshi R, Khetarpal N, Fatokun O, Doucet D, Zubkowski A, Rayat DY, Jackson H, Harford K, Anjum A, Zakir M, Wang F, Tian S, Lee B, Liigand J, Peters H, Wang RQR, Nguyen T, So D, Sharp M, da Silva R, Gabriel C, Scantlebury J, Jasinski M, Ackerman D, Jewison T, Sajed T, Gautam V, Wishart D (2024) DrugBank 6.0: The DrugBank Knowledgebase for 2024. Nucleic Acids Res 52(D1):D1265\u0026ndash;D1275. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/nar/gkad976\u003c/span\u003e\u003cspan address=\"10.1093/nar/gkad976\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNogawa H, Kawai T (2014) hERG Trafficking Inhibition in Drug-Induced Lethal Cardiac Arrhythmia. Eur J Pharmacol 741:336\u0026ndash;339. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ejphar.2014.06.044\u003c/span\u003e\u003cspan address=\"10.1016/j.ejphar.2014.06.044\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNjoku DB, Drug-Induced, Hepatotoxicity (2014) Metabolic, Genetic and Immunological Basis. Int J Mol Sci 15(4):6990\u0026ndash;7003. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ijms15046990\u003c/span\u003e\u003cspan address=\"10.3390/ijms15046990\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMartins V, Ward G, Smyth T, Reader M, Chessari G, Johnson C, Wilsher N (2024) The preclinical pharmacokinetics of Tolinapant-A dual cIAP1/XIAP antagonist with in vivo efficacy. Pharmacol Res Perspect 12:e70030. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/prp2.70030\u003c/span\u003e\u003cspan address=\"10.1002/prp2.70030\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHubatsch I, Ragnarsson EGE, Artursson P (2007) Determination of Drug Permeability and Prediction of Drug Absorption in Caco-2 Monolayers. Nat Protoc 2(9):2111\u0026ndash;2119. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nprot.2007.303\u003c/span\u003e\u003cspan address=\"10.1038/nprot.2007.303\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\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":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-cheminformatics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"chin","sideBox":"Learn more about [Journal of Cheminformatics](https://jcheminf.biomedcentral.com/)","snPcode":"13321","submissionUrl":"https://submission.nature.com/new-submission/13321/3","title":"Journal of Cheminformatics","twitterHandle":"@jcheminf","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Pretraining, ADMET, XLNet, RNN, Molecular Fingerprints","lastPublishedDoi":"10.21203/rs.3.rs-8326331/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8326331/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties of drugs are critical to their efficacy and safety in clinical trials; however, traditional machine learning methods have limited generalization ability in ADMET prediction due to insufficient data. To address this issue, we developed DCPM-ADMET, an innovative pre-trained model with higher accuracy, whose architecture employs a two-channel system\u0026mdash;including an XLNet-based module for capturing the semantics of molecular sequences, an RNN-based component for small molecule property extraction, and ECFP fingerprints for capturing molecular substructures\u0026mdash;and after initial pre-training, the model outperforms traditional methods in prediction accuracy on multiple benchmark datasets for molecular properties; additionally, we fine-tuned it on a self-constructed database containing 465,470 entries covering 97 ADMET properties, and by integrating these 97 prediction models and 36 computational properties, we further developed a free online ADMET prediction tool with 133 endpoints (available at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://admet.bioai-global.com/\u003c/span\u003e\u003cspan address=\"http://admet.bioai-global.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), which is designed to assist researchers in conducting comprehensive molecular ADMET predictions.\u003c/p\u003e \u003cp\u003eScientific contribution\u003c/p\u003e \u003cp\u003eThe development of DCPM-ADMET represents a seminal advancement in computational pharmacology. This novel pre-trained model successfully addresses the fundamental limitation of poor generalization in predicting ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties, a challenge stemming from data insufficiency in traditional machine learning approaches. Our architecture innovatively employs a dual-channel system: an XLNet-based module for deep capture of molecular sequence semantics, an RNN-based component for efficient extraction of small molecule properties, and ECFP fingerprints to comprehensively encode structural features. Following intensive pre-training, DCPM-ADMET demonstrates superior predictive accuracy across multiple benchmark molecular property datasets. Furthermore, we fine-tuned this model on a proprietary, large-scale database of 465,470 entries covering 97 ADMET endpoints. By integrating the resultant 97 prediction models with 36 calculated physicochemical properties, we have deployed a free, high-throughput online ADMET prediction tool with 133 endpoints, which is set to become an essential resource for guiding early-stage drug discovery and safety assessment.\u003c/p\u003e","manuscriptTitle":"DCPM-ADMET: Fusion of Dual-channel Pre-trained Model and Molecular Fingerprints to enhance Drug ADMET Properties Prediction","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-19 19:07:57","doi":"10.21203/rs.3.rs-8326331/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-30T04:57:07+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-25T08:54:43+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-19T07:16:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"171202971071403392406571402432060899714","date":"2026-03-17T01:00:04+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"246510712503360191865935327364760289898","date":"2026-02-24T05:14:27+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-12-17T22:21:27+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-12-11T12:39:52+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-12-11T12:35:48+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Cheminformatics","date":"2025-12-10T10:31:32+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-cheminformatics","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"chin","sideBox":"Learn more about [Journal of Cheminformatics](https://jcheminf.biomedcentral.com/)","snPcode":"13321","submissionUrl":"https://submission.nature.com/new-submission/13321/3","title":"Journal of Cheminformatics","twitterHandle":"@jcheminf","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"cca5aca5-8d1b-4870-b69c-a55601bb9efe","owner":[],"postedDate":"December 19th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-18T02:38:17+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-19 19:07:57","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8326331","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8326331","identity":"rs-8326331","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}