A Novel Machine Learning-enhanced Microfluidic CircRNAs Detection Platform for Breast Cancer Precision Diagnosis

A Novel Machine Learning-enhanced Microfluidic CircRNAs Detection Platform for Breast Cancer Precision Diagnosis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article A Novel Machine Learning-enhanced Microfluidic CircRNAs Detection Platform for Breast Cancer Precision Diagnosis Lei Mou, Xinyu Zhang, Zixin Lin, Yongmei Chen, Huanyu Zhou, yuetao zhang, and 6 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6504183/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The discovery and favourable detection of breast cancer biomarkers are significant for cancer diagnosis. Here, we show that hsa_circ_044235 and hsa_circ_000250 are effective biomarkers for breast cancer diagnosis. Moreover, we present an integrated electrochemical microfluidic circRNAs detection platform (ECMCDP) that combined gold platinum nanoparticles (AuPts)-modified screen-printed electrodes (SPEs), catalytic hairpin assembly (CHA), electrochemical microfluidic chip and a customed low-power electronic system for simultaneous detection of two breast cancer-associated circRNAs. The limit of detection (LOD) were 0.12 fM and 0.1 fM, and the diagnostic accuracy were 92.50% and 88.75% in clinical blood samples, respectively. The platform was validated using paired pre-/post-operative blood and tissue samples. Combined with five machine learning-based diagnostic models, the ensemble diagnosis model achieved a high accuracy of 93.75%. This work aims to identify novel breast cancer biomarkers and establish an innovative circRNAs detection platform to improve breast cancer diagnosis and support clinical prognosis assessment. Physical sciences/Nanoscience and technology/Nanobiotechnology/Biosensors Biological sciences/Cancer/Breast cancer Physical sciences/Nanoscience and technology/Nanobiotechnology/Microfluidics Breast cancer CircRNAs Electrochemical biosensor Point-of-Care detection Machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Breast cancer (BC) is the most commonly diagnosed cancer and the leading cause of cancer death among women in 2022 1,2 . With the in-depth study of the molecular characteristics of BC, the diagnosis and classification methods of BC have changed significantly. New biomarkers are essential to improve diagnostic accuracy, early detection, and personalized treatment 3 , 4 . Circular RNA (circRNA) are exonuclease-resistant, covalently closed RNAs generated via backsplicing, lacking terminal structures (5′ caps/3′ poly-A tails). Their stability and ubiquitous presence in bodily fluids highlight their potential as cancer biomarkers for diagnostic and therapeutic monitoring 5 – 7 . Therefore, the discovery of circRNAs closely related to BC and the realization of its highly sensitive and specific detection are essential for the early diagnosis of BC. However, the existing detection methods have limitations such as cumbersome operation, long time, high background signal, and complex experiment, which hinder the rapid clinical detection application 8 – 10 . It is urgent to develop simple, economical, high-sensitivity, and high-specificity detection methods. Isothermal nucleic acid amplification is a promising nucleic acid detection method that can detect DNA, RNA, cells, and other biological targets. It has the advantages of rapid, low cost, simple operation and is suitable for field detection 11 – 13 . The technique includes enzyme-assisted amplification and enzyme-free amplification. Enzyme-assisted amplification methods such as rolling ring amplification (RCA) and loop-mediated isothermal amplification (LAMP) have disadvantages such as high cost, difficult storage, and complex reaction conditions 14 . Compared with other enzyme-free amplification methods, including hybrid chain reaction (HCR), DNAzyme, and entropy-driven circuit (EDC), catalytic hairpin assembly (CHA) has a simpler and more stable reaction system and higher catalytic efficiency 15 . Electrochemical biosensors combine electronic transduction systems with biorecognition elements, achieving sensitive signal detection through electrochemical methods and molecular specificity via selective bioaffinity interactions 16 – 18 . Surface amplification of nucleic acids enables the integration of electrochemical biosensors and targeted amplification methods on the same platform. This integration reduces assay time, reduces the risk of contamination, and also allows point-of-care (POC) molecular detection 19 , 20 . The electrochemical conversion strategies are classified into labeling-free (studying the rate of electron transfer between redox pairs and modified electrode surfaces in solution) and labeling-based (using redox substances such as methylene blue (MB) and ferrocene (Fc)) methods. These substances bind to the generated long single-stranded DNA (ssDNA) 21 . ssDNA carries many negative charges, and the sequence can be designed to be an excellent biopolymer. Whose assembly is driven by metal ion-DNA coordination and electrostatic force 22 . Various noble metal nanoparticles, like gold nanoparticles (AuNPs) and platinum nanoparticles (PtNPs), improve current response due to their conductivity and catalytic properties 23 , 24 . Moreover, AuNPs and PtNPs can bind to thiol-modified ssDNA 25 . Pt-S bonds are more stable than Au-S bonds. Adding Pt to AuNPs forms stable bioconjugates and reduces interference from biothiols 26 , 27 . Compared with single-metal nanoparticles, bimetallic nanoparticles are more stable and biocompatible with specific optical or electrical activities 28 . In this study, we discovered two BC-related circRNAs and established a new highly sensitive circRNAs detection platform. Circular RNA microarray technology was used to analyze circRNAs expression profiles in BC serum. The value of hsa_circ_044235 and hsa_circ_000250 in the diagnosis of BC was verified by digital PCR (ddPCR) and real-time fluorescent quantitative reverse transcription polymerase chain reaction (qRT-PCR) (Fig. 1 a). Traditional qRT-PCR is complex and subject to contamination risks, microfluidic integrated isothermal amplification technology provides a portable, automated solution for POC detection and cancer screening 29 – 32 . Further, we developed a highly sensitive and specific electrochemical microfluidic circRNA detection platform, used gold platinum nanoparticles (AuPts)-modified screen-printed electrodes (SPEs), catalytic hairpin assembly (CHA), and a custom-designed low-power electronics system (Fig. 1 b). It was successfully applied to the detection of clinical blood and tissue samples (Fig. 1 c). Machine Learning (ML) has significantly improved the accuracy and efficiency of tumour diagnosis, enabling early detection and personalized treatment by analyzing medical imaging and molecular data, and promoting multimodal data integration and clinical translation 33 , 34 . This study used five machine learning models to achieve efficient, accurate, and economical breast cancer detection and prediction, and the accuracy of the ensemble model was 93.75%. Additionally, it offers a theoretical and practical basis for identifying other disease biomarkers. 2. Results 2.1 Discovery of new biomarkers for BC diagnosis To explore the difference in serum circRNAs expression between the healthy control group (HCs) and BC patients group. We randomly collected serum samples from 3 HCs without cancer and 3 patients with BC. Arraystar human circRNAs microarray technology was used to sequence 13,433 circRNAs in serum, of which 7329 were up-regulated, and 614 were down-regulated. According to P 1.5, a total of 105 differentially expressed circRNAs were identified, including 54 up-regulated circRNAs and 51 down-regulated circRNAs. Significantly differentially expressed circRNAs were selected ( Supplementary Fig. 1a, Supplementary Table 1 ). The heat map showed the top 50 differentially expressed circRNAs (Fig. 2 a). To verify the accuracy of circRNAs preliminary screening results, we selected three up-regulated circRNAs (hsa_circ_102101, hsa_circ_104293, hsa_circ_400241) and down-regulated circRNAs (hsa_circ_008053, hsa_circ_008016, hsa_circ_103637) were verified by qRT-PCR. The qRT-PCR results were consistent with the gene sequencing results (Fig. 2 b, c). The results showed that the sequencing results were accurate and reliable. Based on its most significant differential expression profile in serum (FC > 2.99, P < 0.05), hsa_circ_044235 was prioritized for subsequent studies. Sequencing results showed that hsa_circ_000250 was also differentially expressed in BC patients ( Supplementary Fig. 1b ). The qRT-PCR results showed that hsa_circ_044235 expression decreased in serum of BC patients compared with HCs (Fig. 2 d). Meanwhile, we constructed a ddPCR method for the detection of hsa_circ_044235. Explored the optimal annealing temperature and total sample size and selected 42.8°C as the optimal annealing temperature ( Supplementary Fig. 2a, b ) and 5 ng as the optimal total sample size ( Supplementary Fig. 2c, d ). ddPCR showed decreased expression of hsa_circ_044235 in the serum of BC patients compared with HCs (Fig. 2 e). qRT-PCR results showed that hsa_circ_000250 was increased in the serum of BC patients compared with HCs (Fig. 2 f). For all HCs and BC patients, the area under the curve (AUC) values of hsa_circ_044235 and hsa_circ_000250 were 0.87 (95% confidence interval (CI) 0.83–0.91) and 0.88 (95% CI 0.81–0.95) respectively (Fig. 2 g). Furthermore, compared with linear RNA and human-β-actin, circRNA exhibited higher temporal stability and better resistance to enzymatic degradation (Fig. 2 h, i). These results indicated that hsa_circ_044235 and hsa_circ_000250 show significant expression differences in BC patients, and combined with their high diagnostic sensitivity and specificity, they can be used as effective biomarkers for breast cancer diagnosis. 2.2 Preparation and characterization of electrochemical biosensors The electrochemical biosensor was fabricated using a multilayer screen printing method, with conductive ink, Ag/AgCl reference ink, and UV insulating ink printed on polyethylene glycol terephthalate (PET) film, respectively ( Supplementary Fig. 3 ). Before the sensor was functionalized, we activated the SPEs using the oxygen plasma method ( Supplementary Fig. 4a ). Explored the plasma treatment time, the SPEs surface current value achieved maximum when plasma treatment time was 6 min ( Supplementary Fig. 4b, c ). The electrical conductivity of the SPEs was significantly increased by plasma activation ( Supplementary Fig. 4d ). SPEs were hydrophobic with a contact Angle (CA) of 110 ( Supplementary Fig. 5a ). After plasma, the SPEs exhibited a cleaner surface with increased hydrophilicity (CA = 39.52) ( Supplementary Fig. 5b ).AuNPs and PtNPs were deposited in situ on the SPEs by cyclic voltammetry (CV) method to increase the electrochemical performance of the SPEs ( Supplementary Fig. 6 ). Moreover, the concentration, deposition cycles and ratio of Au:Pt were optimized. The current value of the SPEs surface achieved the maximum under the optimized conditions. (The concentrations of HAuCl 4 ·3H 2 O and K 2 PtCl 4 were 1 mM and 5 mM, respectively, ( Supplementary Fig. 7 ), with 15 deposition cycles ( Supplementary Fig. 8 ) and an Au:Pt ratio of 1:1 ( Supplementary Fig. 9a )). Meanwhile, the biosensor response to hsa_circ_000250 was maximum when the Au:Pt ratio was 1:1 ( Supplementary Fig. 9b ). The CA on the SPEs surface after AuPt deposition was 68.20 ( Supplementary Fig. 5c ). AuPt was deposited on SPEs in situ (Fig. 3 a). The morphology and chemical composition of AuPt-SPEs were characterized. Compared with bare SPEs, SEM showed that AuPt nanoparticles were successfully and evenly distributed on the SPE surface (Fig. 3 b, c). The successful fusion of AuNPs with PtNPs nanoparticles were confirmed by high-resolution TEM imaging, which showed lattice stripes in different states of Au and Pt ( Supplementary Fig. 10 , Fig. 3 d). The diameter distribution of AuPt nanoparticles ranged from 20 to 80 nm (Fig. 3 e). EDX and EDS analysis further demonstrated the successful fusion of AuNPs and PtNPs (Fig. 3 f, g). AFM was used to characterize the average surface roughness of bare SPEs and AuPt-SPEs. Compared with the bare SPE, the surface roughness of AuPt-SPEs is significantly increased (Fig. 3 h, i). After AuPt was confirmed to be successfully deposited on the SPE surface. CV was used to test the electrochemical surface active area of the bare SPEs and the AuPt-SPEs, which were 0.006 cm 2 and 0.037 cm 2 , respectively. Compared with the bare SPEs, the electrochemical surface active area of AuPt-SPEs increased by nearly 5 times. It is further indicated that the deposition of AuPt on the SPEs surface significantly increases the electrochemical active sites and improves the electrochemical performance of the SPEs. 2.3 Feasibility of electrochemical biosensors The design of the electrochemical biosensor is based on attaching a thiolated hairpin probe (SH-H1/H1') to the AuPt-SPEs surface. The target circRNA can open and bind to the H1/H1' hairpin structure. Another ferrocene-labeled hairpin probe (Fc-H2/H2') can form a double-stranded structure with the open H1/H1' and gradually replace the release target circRNA. The Fc-H2/H2' is close to the AuPt-SPEs. The detected electrochemical signal is proportional to the target circRNA content (Fig. 4 a). The feasibility of our electrochemical biosensor was initially demonstrated using native polyacrylamide gel electrophoresis (Native-PAGE). Only one band was visible when H1/H1', H2/H2', or circRNA was present alone. There are two separate bands in the presence of both H1/H1' and H2/H2', proving that no hybridization between H1 and H2 occurs without the target circRNA. A new band with slow electrophoretic mobility and large molecular weight appeared after the addition of the target circRNA, indicating that the addition of the target circRNA triggered the CHA process of H1 and H2 (Fig. 4 b, c). Further, the ΔG of H1-H2, H1'-H2' combined is smaller than that of H1/H1', H2/H2', and H1/H1'-circRNA, demonstrating the feasibility of electrochemical biosensors 35 ( Supplementary Fig. 11, 12 ). CV and open circuit potential-electrochemical impedance spectroscopy (OCP-EIS) were used to further characterize the surface of AuPt-SPEs after each modification step (Fig. 4 d, e). When AuPt original was deposited on the surface of SPEs, the peak current height increased in the CV diagram, and the resistance decreased in the Nyquist diagram. These changes indicated that the conductivity and electrochemical catalytic activity of the SPEs increased. Subsequent modifications on the surface of the AuPt-SPEs prevent charge transfer on the electrode surface, resulting in lower peak current height in the CV diagram and higher resistance in the Nyquist diagram. After confirming the successful modification of AuPt-SPEs, the probe concentration, reaction time, and reaction temperature were further optimized to achieve the best detection performance ( Supplementary Fig. 13, 14 ). Both circRNAs showed the best detection performance under the condition that the probe concentration was 2 µM and the reaction time was 37°C for 2 h. The electrochemical biosensor performance was evaluated by differential pulse voltammetry (DPV) (Fig. 4 f, g). The electrochemical biosensor showed a log-linear relationship between the peak current height of DPV and the target concentration. The ultra-low LOD of hsa_circ_004423 and hsa_circ_000250 detected by electrochemical biosensors were 0.12 fM and 0.01 fM, respectively (Fig. 4 h, i). Stability was assessed by testing the DPV response of AuPt-SPEs placed at 4°C from 1 to 14 days, with a relative standard deviation (RSD) of 4.44% at 14 days (Fig. 4 j). Five SPEs were prepared in parallel to detect different concentrations of circRNA (100 nM, 10 nM) to evaluate the reproducibility of the electrochemical biosensor, with RSD of 3.91% and 5.52% for the two concentrations, respectively (Fig. 4 k). The electrochemical biosensor showed high selectivity in the presence of high concentrations of interfering substances (Fig. 3 l, n). These results verified the feasibility of the reaction system and proved that the biosensor has remarkable stability, reproducibility and specificity. Demonstrated its ability to meet actual testing requirements. 2.4 Design and fabrication of ECMCDP To realize low-cost, simple and integrated portable detection, this study designed a compact wireless device integrating a microfluidic chip and a flexible printed circuit board (FPCB). The results were compared with those of a commercial electrochemical workstation (Fig. 5 a, Supplementary Fig. 15a) . The microfluidic chip consists of a microchannel and microchamber layer, an electrode layer, VHB glue, and a PDMS support layer (Fig. 5 b). The microfluidic chip comprised a 10 µL H1/H1' prestorage chamber, a 10 µL H2/H2' prestorage chamber, a 20 µL sample RNA reservoir, a mixing channel, an electrochemical reaction zone, a 100 µL wash chamber, and a 200 µL waste collection chamber. Under the action of a micro-peristaltic pump, the reagents are transported from different storage areas through the mixing channel to the reaction area, triggering the CHA reaction with H1/H1' fixed on the AuPt-SPEs. ( Supplementary Fig. 15b ). Using COMSOL software to simulate the mixing effect of mixing channels. In the designed mixing channel, two liquids of different concentrations mix well and reach the reaction zone at a constant speed (Fig. 5 c). Since the CHA reaction required 2 hours at 37°C, the stable liquid volume in the microfluidic system confirmed effective sealing and compatibility with optimal reaction conditions. ( Supplementary Fig. 16 ). Red and blue inks simulated the reaction process, with reagents preloaded in storage chambers. After the micro peristaltic pump was activated, liquids flowed through the mixing channel into the reaction zone. Post-reaction, a washing solution was introduced to quench the reaction (Fig. 5 d). The microfluidic system was connected to a reusable PCB potentiostat system. The PCB potentiostat system block diagram (Fig. 5 e) and circuit diagram ( Supplementary Fig. 17 ) were used to measure the electrical response signal during the process. In the MCU-based FPCB control system, the voltage output is programmed by the integrated DAC, and the potentiostat module and electrode selection module drive the sensor electrodes. The current signal of the impedance amplifier (precision operational amplifier with analog switches) was collected by the ADC and transmitted to the smartphone via Bluetooth for real-time monitoring and analysis. The DPV results of ECMCDP were compared with those of commercial electrochemical workstations. These results indicated the reliability of the output results of ECMCDP (Fig. 5 f). The microfluidic system costs only $ 2.08, and the FPCB can be reused ( Supplementary Table 2 ). It can provide an economical and feasible technical scheme for large-scale BC screening programs. 2.5 Clinical feasibility of ECMCDP To preliminarily assess the potential clinical application of ECMCDP in BC diagnosis, blood samples were collected from 44 BC patients and 36 HCs. Additionally, we obtained surgically - removed normal, para-carcinoma, and carcinoma tissues from 7 BC patients. The results of ECMCDP were compared with those of traditional qRT-PCR. Compared with traditional qRT-PCR, which requires a complex operation process such as reverse transcription, ECMCDP can directly detect total RNA and output the results (Fig. 6 a). ECMCDP successfully detected hsa_circ_044235 and hsa_circ_000250 in the blood of HCs and BC patients. The expression trends of the two circRNAs were consistent with our previous verification results (Fig. 6 b, c). Meanwhile, correlation analysis was conducted between the current values of the two circRNAs and Ct values ( Supplementary Table 3 ). It can be seen from the figure that the changes between the current values of the two cricRNAs and Ct values were consistent and showed a significant negative correlation. Spearman's r values were − 0.6256 and − 0.5942, respectively (Supplementary Fig. 18) . This indicates that the higher the Ct value, the lower the circRNAs expression level and the corresponding current value. Moreover, ECMCDP, even at high Ct values (> 30), can also detect the current signal. ROC curves of ECMCDP detection were plotted, and the AUC were calculated to discuss the diagnostic accuracy of the ECMCDP (Fig. 6 d, Supplementary Table 4 ). The AUC values of hsa_circ_044235 and hsa_circ_000250 were 0.98 (95% CI 0.94–1.00) and 0.92 (95% CI 0.81–1.00), respectively. Compared with other commonly used clinical protein biomarkers, the AUC of CEA, CA125, and CA153 were respectively 0.86 (95% CI 0.74–0.98), 0.82 (95% CI 0.70–0.95) and 0.67 (95% CI 0.50–0.85). Results from the study samples of HCs and BC patients demonstrate the potential of these two circRNAs as diagnostic biomarkers for BC. They also showed the practical application of the ECMCDP for the early diagnosis of BC. The AUC of the combined detection of two circRNAs and other protein biomarkers was 1.00, indicating that the collaborative detection between biomarkers can significantly improve the diagnostic efficiency of BC (Supplementary Table 5) . The ROC curves provided cut-off current values of 0.17 and 0.19 for the sum of optimized sensitivity and specificity. Using these cut-off values, ECMCDP can accurately distinguish positive and negative clinical samples with an accuracy of 92.50% and 88.75%, respectively, with high sensitivity and specificity (Fig. 6 e, Supplementary Table 6 ). ECMCDP stands out for its high sensitivity and accuracy compared to other methods for detecting breast cancer-related biomarkers ( Supplementary Table 7 ). Monitoring biomarker levels after treatment can indicate response evaluation, predict disease prognosis, and guide subsequent individualized therapy. Therefore, we collected preoperative and postoperative blood samples from 4 BC patients and used ECMCDP to detect the changes of two circRNAs before and after the operation. Compared with the preoperative results, the expression of hsa_circ_044235 in the blood of BC patients increased, and hsa_circ_000250 decreased, indicating their potential as effective monitoring biomarkers of BC (Fig. 6 f, g, Supplementary Table 8 ). ECMCDP was also used to detect the expression of two circRNAs in normal tissues, para-carcinoma tissues, and carcinoma tissues of 7 BC patients. Compared with normal and para-carcinoma tissues, hsa_circ_044235 was up-regulated, and hsa_circ_000250 was down-regulated in carcinoma tissues. Contrary to the trend of expression in blood, this suggests the possibility of both circRNAs as BC tissue-specific biomarkers (Fig. 6 h, i, Supplement Table 9a, b ). The ECMCDP demonstrated high specificity and sensitivity in detecting hsa_circ_004423 and hsa_circ_000250 across diverse human biological samples, indicating broad clinical utility. 2.6 ML for BC diagnosis Gradient boosting (GB), support vector machine (SVM), random forest (RF), and logistic regression (LR) models were used to process the current signals of hsa_circ_044235 and hsa_circ_000250 derived from ECMCDP. Soft voting was used to predict the diagnosis of BC (Fig. 7 a). The chord chart showed the association of biomarkers with patients (bandwidth = relative contribution). It can be seen that circRNA markers are mainly closely associated with BC patient samples, and hsa_circ_000250 is weaker than hsa_circ_044235 with patients (Fig. 7 b). Principal component analysis (PCA) showed that BC patients (n = 44) and HCs (n = 36) were clearly separated in 2D PCA space (Fig. 7 c). Correlation heatmaps for all biochemical markers, including circRNAs and traditional biomarkers, revealed heterogeneity among patients and potential typing trends through Pearson correlation coefficients (Fig. 7 d). The Pearson correlation heatmap between the biochemical markers showed the correlation coefficients of their expression change trends across all samples, highlighting potential synergistic or complementary relationships (Fig. 7 e). Feature importance analysis of the model highlighted hsa_circ_044235 and hsa_circ_000250 as the most contributing features in the model (Fig. 7 f). SVW, LR, RF, and GB were used to analyze the true positive (TP), false positive (FP), true negative (TN), and false negative (FN) (Fig. 7 g) of the ensemble model on the validation set (20% data). Five ML models achieved BC classification accuracies of 86.67%, 87.50%, 93.75%, 100%, and 100%, with the ensemble model reached 93.75% accuracy (Fig. 7 h). Confusion matrix analysis of the validation set showed high sensitivity and specificity for correct classification of patients and healthy individuals (Fig. 7 i). Comparison of the ROC curves of the deep neural network model and the soft voting ensemble model on the validation set showed that the ensemble model had robust diagnostic performance with an AUC of 0.94 (Fig. 7 j). Therefore, ML provides a robust and scalable computational framework to systematically evaluate the performance of enhanced ECMCDP in detecting circRNAs. 3. Conclusion In this study, we identified two novel circRNAs (hsa_circ_044235 and hsa_circ_000250) as highly sensitive and specific serum biomarkers for BC diagnosis. Through comprehensive circRNA microarray screening and subsequent validation via qRT-PCR and ddPCR, we demonstrated that these circRNAs exhibit significant differential expression between BC patients and HCs, with high diagnostic accuracy (AUC = 0.87, 0.88, respectively). Leveraging their stability and resistance to enzymatic degradation, we developed an ultrasensitive electrochemical biosensor based on CHA, achieving femtomolar-level detection limits (0.12 fM and 0.01 fM, respectively). To enable point-of-care testing, we designed an integrated, low-cost electrochemical microfluidic circRNA detection platform, which combines a microfluidic chip with a portable potentiostat system for automated, high-throughput analysis. Clinical validation in 44 BC patients and 36 HCs confirmed the platform’s diagnostic efficacy, with AUC values of 0.98 and 0.92 for hsa_circ_044235 and hsa_circ_000250, respectively-outperforming conventional protein biomarkers (CEA, CA125, CA153). Furthermore, this platform demonstrated excellent reproducibility, stability, clinical feasibility in monitoring postoperative treatment responses and distinguishing cancerous from non-cancerous tissues. The integration of ML enhanced diagnostic precision, with ensemble models (GB, SVM, RF, LR, Neural Network) achieving 93.75% accuracy in classifying BC patients. Feature importance analysis confirmed that the two circRNAs were the most predictive biomarkers, reinforcing their clinical utility. Additionally, this platform successfully monitored postoperative biomarker dynamics, suggesting its applicability in treatment response assessment. The discovery of circRNA biomarkers and their integration into an ML-powered microfluidic platform paves the way for large-scale, early-stage BC screening, with potential extensions to other cancers. Future studies should explore the mechanistic roles of these circRNAs in BC progression and validate this platform in larger, multi-center cohorts. This work represents a significant advancement in liquid biopsy-based cancer diagnostics, offering a transformative approach to precision oncology through biomarker discovery, biosensor engineering, and ML-driven data analysis. 4. Methods 4.1 Materials and reagents All DNA and circRNAs were custom-synthesized by Sangon Biotechnology Co., Ltd. (Shanghai, China) and dissolved in a DEPC-treated aqueous solution. The RNA solution can be used directly without treatment ( Supplementary Table 10 ). DNA solutions were formed into hairpin structures by heating at 95°C for 5 min followed by slow cooling at 6°C/min to room temperature. Tris (2-carboxyethyl) phosphine hydrochloride (TCEP) was purchased from MedChemExpress. 6-sulfhydryl-1-hexanol (MCH) and gold (III) chloride trihydrate (HAuCl 4 ·3H 2 O) were purchased from Sigma-Aldrich. Potassium chlorite (K 2 PtCl 4 ) was purchased from Aladdin. DNA-free ribonuclease water was purchased from Yitao Biotechnology Co., Ltd. (Guangzhou, China). Carbon ink (CH-8) and UV ink were purchased from China Pinghu Juju Consumable Technology (Pinghu) Co., Ltd. Ag/AgCl ink (011464) was purchased from ALS, Tokyo, Japan. The wash solution contained 10 mM Tris-Hcl, 50 mM NaCl, and 0.05% TWeen-20. RNA extraction kit, reverse transcription kit, and SYBR Green Ⅰ were purchased from Accurate Biology Co., Ltd. (Hunan, China). Digital PCR probe method supermix, probe method microdroplet generation oil, and microdroplet reading oil were purchased from Bio-rad, USA. The polydimethylsiloxane (PDMS; Sylgard 184) kit was purchased from Dow Corning. 4.2 Instruments CircRNAs in serum were sequenced using Arraystar human circRNAs microarray technology (China Kangcheng Biotechnology Co., Ltd.). ddPCR was performed using a Bio-Rad QX200 droplet digital PCR system, USA. qRT-PCR was performed using the QuantStudio 3 real-time PCR System. Native-PAGE was performed using the Bio-Rad Mini-PROTEAN Tetra System, USA. Electrochemical characterization, including DPV, CV, OCP-EIS, was performed on a CHI650E electrochemical workstation (Shanghai Chenhua Instrument Co., Ltd., China). Plasma instruments were purchased from Zhongxinqinheng Co., Ltd. (Suzhou, China). The morphology of AuPt, as well as its dimensions, were characterized using a scanning electrode (Thermo Fisher TF20) and a transmission electron microscope (Zeiss Gemini SEM360). The surface roughness was analyzed by atomic force microscopy (NT-MDT, Russia). A miniature peristaltic pump (BW100-WP110) was purchased from Baoding Chuangrui Pump Co., Ltd. 4.3 Clinical blood and tissue samples collection The blood and tissue samples of surgically resected BC patients diagnosed by pathology in the Third Affiliated Hospital of Guangzhou Medical University from September 2024 to March 2025 were collected. The blood samples of healthy people were collected from the Physical Examination Center of the Third Affiliated Hospital of Guangzhou Medical University. After collection and repacking, all samples were stored at -80°C. This study was approved by the Ethics Committee of the Third Affiliated Hospital of Guangzhou Medical University (2025048). 4.4 Serum circRNAs sequencing analysis In order to explore the difference in the expression of circRNAs in the serum of HCs group and patient group. We randomly collected serum samples from 3 non-cancer healthy individuals and 3 BC patients. Arraystar human circRNAs microarray technology was used to sequence circRNAs in serum. The FC > 1.5 and P < 0.05 circRNAs were used as significantly differentially expressed circRNAs. 4.5 Preliminary validation of clinical samples qRT-PCR verification of samples after sample collection, total RNA was extracted using an RNA extraction kit (AG21024). After reverse transcription (AG11728), 20 µL of the reaction system was prepared according to the qRT-PCR reagent instructions (AG11718). After mixing, the samples were detected using QuantStudio 3 real-time PCR system. For ddPCR verification, the samples were prepared into a 20 µL ddPCR reaction system on ice. Oil was generated by adding micro-droplets. The micro-droplets were covered with adhesive strips and placed in the micro-droplet generator to form water-in-oil micro-droplets. Four micro-droplets were transferred to a 96-well plate and sealed with a heat-sealing membrane. Amplification was performed using a Bio-Rad T100 gradient PCR apparatus. After amplification, the samples were detected and analyzed by a QX200 droplet digital PCR instrument. 4.6 Preparation of AuPt-SPEs SPEs were designed using Auto CAD 2024 software. hsa_circ _044235 and hsa_circ _000250 shared a reference electrode and counter electrode, and conductive ink CH-8 was printed on the PET film to form the electrode circuit. After printing, it was oven-dried at 80°C for 20 min, followed by screen printing using UV insulating ink and exposing it to UV light for 5 min. The reference electrode was made by directly printing Ag/AgCl ink and allowing it to dry at room temperature. Subsequently, a UV insulating layer was applied to the SPEs surface to expose only the working area and electrode wire contacts. After these steps, 100 W plasma was used for 6 min to activate the SPEs. 1 mM HAuCl 4 ·3H 2 O and 5 mM K 2 PtCl 4 solution in 0.5 M H 2 SO 4 were mixed 1:1 before use. The AuPt composite material was prepared by CV to improve the conductivity of the SPEs, and the scan rate was 0.02 V/s in -0.2 to 0.5 V for 15 cycles. Finally, the SPEs surface was rinsed with deionized water and dried at room temperature. The samples were stored at 4°C. 4.7 Preparation of biosensors The 2 µM thiolated capture probe H1/H1' was incubated with 1 mM TCEP for 30 min at room temperature in the dark to break with disulphide bonds. 5 µL of 2 µM of pre-sulfide-capture probe H1/H1' was dropped onto the surface of the pre-treated AuPt-SPEs and incubated overnight at 4°C. After washing with washing solution, 5 µL of 1 mM MCH was drip-added to the SPEs surface and incubated at 4°C for 1 h to obtain a neatly arranged DNA monolayer, and nonspecific binding sites were blocked. After washing and drying at room temperature, electrochemical biosensors were obtained. Subsequently, 5 µL of 2 mΜ Fc-labeled H2/H2' and 5 µL of hsa_circ _044235 and hsa_circ _000250 were dropped-added to the SPEs surface for CHA amplification. DPV measurements were performed in 10 mM tris-HCl (pH 7.4) to obtain a DPV response signal. The measurement parameters were set as a scan rate of 0.02 V/s, a resting time of 20 s, and a scan voltage range of 0.2 to 0.6 V. 4.8 Electrochemical characterization of AuPt-SPEs For detailed analysis of the surface at each stage of the electrochemically modified SPEs, the CV and OCP-EIS methods were used in 0.5 M KCl solution containing 10 mM K 4 Fe(CN) 6 /K 3 Fe(CN) 6 (1:1). The voltage range was − 0.2 to 1.0 V, the scan rate was 0.02 V/s, 2 cycles, the sampling interval was 0.001 V, and the standing time was 20 s. OCP-EIS was performed with the following Settings: the initial potential was set to the measured OCP voltage with a frequency range of 0.1 to 100,000 Hz and an amplitude of 0.005 V. The Randles-Sevcik equation was used to calculate the active surface area of the SPEs. There was a linear relationship between the CV anode peak current (Ip) and the square root of the scan rate ( \(\:{\text{v}}^{\frac{\text{1}}{\text{2}}}\) ). $$\:\begin{array}{c}\text{I}\text{p}\text{=}\frac{\text{2.69×}{\text{10}}^{\text{5}}\text{×}{\text{n}}^{\frac{\text{3}}{\text{2}}}\text{×C×A×}{\text{v}}^{\frac{\text{1}}{\text{2}}}}{{\text{D}}^{\frac{\text{1}}{\text{2}}}}\#（\text{1}）\end{array}$$ Where Ip refers to the peak current (A), A is the surface active area (cm 2 ), n is the number of transferred electrons (n = 1), C is the concentration of [Fe(CN)] 3−/4− (5×10 − 6 mol/cm 3 ), D is the diffusion coefficient (6.7×10 − 6 cm 2 /s) of [Fe(CN)] 3−/4− in 0.5 M KCI solution, and v is the potential sweep rate (V/s). The surface area of the SPEs was calculated from the slope of Ip vs. \(\:{\text{v}}^{\frac{\text{1}}{\text{2}}}\) . The slope of the bare SPEs and AuPt-SPEs was 20.702 and 128.03, respectively. 4.9 Experimental details of Native -polyacrylamide gel electrophoresis Firstly, H1/H1' and H2/H2' were subjected to CHA reaction with hsa_circ_044235 and hsa_circ_000250 at 37°C for 2 h, respectively. In these reactions, a final concentration of 2 µM (H1/H1', H2/H2') of DNA was used, except for circRNAs, which was 100 nM. The product corresponding to each lane was then mixed with 6×loading buffer. 0.5×Tris-borate-EDTA buffer (TBE) was selected as the running buffer. The electrophoresis experiment was carried out in 15% natural polyacrylamide gel electrophoresis at a constant voltage of 120 V for 2 h, and finally stained with GelstainRed nucleic acid dye. Photographs were taken with a FireReader V10 Plus gel imaging system and finally processed with Photoshop software. 4.10 ECMCDP fabrication and assembly The power supply of the ECMCDP electronic system was regulated by a lithium-ion battery combined with a switched capacitor boost converter (PW510B-3.3V, PWCHIP). The system microcontroller unit (MCU) uses Bluetooth low energy chip NRF52832-QFAA-R (Microchip Technology), and the built-in analog-to-digital converter (ADC) to realize signal acquisition. Voltage programming was generated by the DAC module MCP4725AOT-E/CH (Microchip Technology) and applied to the sensor electrode by the potentiostat module LMV612 (Texas Instruments). The current signal was processed by the trans-impedance amplifier composed of MCP6031T-E/OT precision operational amplifier (Microchip Technology), collected by the ADC of MCU, and transmitted to the smartphone terminal through Bluetooth for real-time analysis. The current signal has its own trans-impedance amplifier channel according to the different detection substances. A microfluidic chip mold for nucleic acid detection integrated with the chip was designed and prepared to form a PMMA mold. The PDMS prepolymer was blended with the curing agent in a 10:1 ratio. The mixture was then poured into the mold and a degassing process was carried out under vacuum for 15 min to eliminate air bubbles. Subsequently, the molds were cured at 85°C for 1 h. The PDMS microfluidic chip was cleaned by soaking in 75% ethanol and deionized water, respectively. The PDMS support layer and the electrode layer were connected with VHB glue and then connected to the FPCB, which was welded to the cell to complete the assembly. 4.11 Detection of ECMCDP in clinical samples After sample collection, total RNA was extracted by RNA extraction kit, and RNA and H2/H2' were directly prestored on the microfluidic chip. Under the action of a peristaltic pump, the reaction was carried out on the surface of H1/H1'-SPEs, and ECMCDP collected the electrochemical signal after 37 ℃ for 2 h. The sensitivity and specificity of the circRNAs detection platform were calculated using the cut-off value of the maximum Youden index (Youden index = maximum value (sensitivity + specificity − 1)). 4.12 Machine Learning All analyses were performed using Python (version 3.5) with publicly available packages, including Pandas (v1.1.5), Scikit-learn (v0.24.2), Keras (v2.4.3), and Tensorflow (v2.4.1). The source code for data preprocessing, model training, and visualization is available under the MIT License at Zenodo ( https://doi.org/10.5281/zenodo.152039487 ) and GitHub ( https://github.com/Lamzsen/BCclassifier , release v1.0.1). 4.13 Statistical analysis Statistical analyses of all data were performed using Graph Pad Prism 9.0 software. Spearman correlation analysis was performed using IBM SPSS Statistics 26. All statistical tests are indicated in the figure legends. LOD was calculated using the standard slope method: $$\:\begin{array}{c}\text{L}\text{O}\text{D}\text{=}\frac{\text{3δ}}{\text{s}}\#（\text{2}）\end{array}$$ Where \(\:\text{δ}\) is the standard deviation of the intercept, and s is the slope of the standard curve. Declarations Supporting Information The Supporting Information is available free of charge at http://. or from the author. Author Information Corresponding Author: * Yong Xia: [email protected] *Lei Mou: [email protected] *Honghai Hong: [email protected] Author Contributions ： Xinyu Zhang: conceptualization, data collation, formal analysis, investigation, validation, writing-first draft, writing-review, and editing. Zixin Lin: Conceptualization, data management, visualization, writing-first draft, writing-review and editing. Yongmei Chen: Data management, formal analysis, visualization, writing-review and editing. Huanyu Zhou: Methodology, writing review and editing. Yuetao Zhang: Formal analysis, software, writing-review and editing. ZiJie Li: data management, formal analysis, writing-review and editing. Runlin lv: Resources, writing-review and editing. Zhiqi Li: Research, writing, review, editing. Rubing Xiong: Research, writing, review, editing. Yong Xia: Conceptualization, writing-review and editing, funding acquisition, supervision. Mou Lei: Conceptualization, writing-review and editing, funding acquisition, supervision. Honghai Hong: Concept, methodology, writing-review and editing, funding acquisition, supervision. Funding Information: We thank Guangzhou Municipal Science and Technology Foundation (2025A03J3757); Department of Science and Technology of Guangdong Province (2023A1515110357); Guangzhou Medical University (02-408-2203-2058, PX-66242684) for financial support. Data availability ： All data supporting the findings of this study are available in the paper and its Supplementary Information. Source data are provided in this paper. Competing Interests: The authors declare no competing financial interests. References Bray, F. et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. Ca-Cancer. J. Clin. 74, 229–263 (2024). Wu, S. et al. V2C-Driven Nanodelivery Platform Potentiates Synergistic Breast Cancer Therapy. ACS. Mater. Lett. 5, 3017–3031 (2023). Loibl, S. et al. Breast cancer. The Lancet. 397, 1750–1769 (2021). Chen, B. et al. Targeting non-coding RNAs to overcome cancer therapy resistance. Signal. Transduct. Tar. 7, 121 (2022). Chen, J. et al. Circular RNA circRHOBTB3 represses metastasis by regulating the HuR-mediated mRNA stability of PTBP1 in colorectal cancer. Theranostics. 11, 7507–7526 (2021). Gong, X. et al. Circular RNA circEsyt2 regulates vascular smooth muscle cell remodeling via splicing regulation. J. Clin. Invest. 131 (2021). Conn, V. M. et al. Circular RNA in cancer. Nat. Rev. Cancer. 24, 597–613 (2024). Zhang, B. et al. An integrated electrochemical POCT platform for ultrasensitive circRNA detection towards hepatocellular carcinoma diagnosis. Biosens. Bioelectron. 192, 113500 (2021). Sha, L. et al. Collaborative CRISPR-Cas System-Enabled Detection of Circulating Circular RNA for Reliable Monitoring of Acute Myocardial Infarction. Small. 20, 2402895 (2024). Digby, B. et al. Computational approaches and challenges in the analysis of circRNA data. Bmc. Genomics. 25, 527 (2024). Qian, J. et al. A microdroplet SERS-RCA biosensor with enhanced specificity and reproducibility for profiling dual miRNAs in idiopathic pulmonary fibrosis diagnosis and monitoring. Chem. Eng. J. 482, 149012 (2024). Mohammadniaei, M. et al. A non-enzymatic, isothermal strand displacement and amplification assay for rapid detection of SARS-CoV-2 RNA. Nat. Commun. 12, 5089 (2021). Zhao, Y. et al. Isothermal Amplification of Nucleic Acids. Chem. Rev. 115, 12491–12545 (2015). Zhang, C. et al. Application of Nanomaterials in Isothermal Nucleic Acid Amplification. Small. 18, 2102711 (2022). Liu, J. et al. Applications of Catalytic Hairpin Assembly Reaction in Biosensing. Small. 15, 1902989 (2019). Zhang, Y. et al. A fully integrated, non-invasive, and multimodal wearable device for sweat stimulation, collection and multiple physiological signals analysis. Chem. Eng. J. 505, 159209 (2025). Ye, C. et al. Nucleic acid-based wearable and implantable electrochemical sensors. Chem. Soc. Rev. 53, 7960–7982 (2024). Zhang, Y. S. et al. Multisensor-integrated organs-on-chips platform for automated and continual in situ monitoring of organoid behaviors. PNAS . 114, E2293-E2302 (2017). Zheng, W. et al. Diagnosis of paediatric tuberculosis by optically detecting two virulence factors on extracellular vesicles in blood samples. Nat. Biomed. Eng. 6, 979–991 (2022). Kashefi-Kheyrabadi, L. et al. Rapid, multiplexed, and nucleic acid amplification-free detection of SARS-CoV-2 RNA using an electrochemical biosensor. Biosens. Bioelectron. 195, 113649 (2022). Dastidar, M. G. et al. Emerging electrochemical nanoscale architectures for point-of-care detection of miRNA biomarkers. Chem. Eng. J. 508, 160903 (2025). Cobzariu, M. A. et al. Enzyme-assisted isothermal amplification of nucleic acids on the electrode surface. Curr. Opin. Electroche. 40, 101322 (2023). Chen, K. et al. A novel signal amplification label based on AuPt alloy nanoparticles supported by high-active carbon for the electrochemical detection of circulating tumor DNA. Anal. Chim. Acta. 1169, 338628 (2021). Liu, Y. et al. Fast and in-situ electrodeposition of MXene/AuNPs composite for multiplexed and sensitive detection of infectious biomarkers using an electrochemical biosensor. Microchem. J. 207, 112064 (2024). Tu, J. et al. A wireless patch for the monitoring of C-reactive protein in sweat. Nat. Biomed. Eng. 7, 1293–1306 (2023). Song, S. et al. Photo-Assisted Robust Anti-Interference Self-Powered Biosensing of MicroRNA Based on Pt–S Bonds and the Inorganic–Organic Hybridization Strategy. Anal. Chem. 94, 1654–1660 (2022). Quan, K. et al. Pt-Decorated Gold Nanoflares for High-Fidelity Phototheranostics: Reducing Side-Effects and Enhancing Cytotoxicity toward Target Cells. Angew. Chem. Int. Edit. 63, e202402881 (2024). Cao, X. et al. Bimetallic AuPt nanochains: Synthesis and their application in electrochemical immunosensor for the detection of carcinoembryonic antigen. Biosens. Bioelectron. 39, 226–230 (2013). Yang, B. et al. Programmable CRISPR-Cas9 microneedle patch for long-term capture and real-time monitoring of universal cell-free DNA. Nat. Commun. 13, 3999 (2022). Liu, Y. et al. Multiplexed discrimination of SARS-CoV-2 variants via plasmonic-enhanced fluorescence in a portable and automated device. Nat. Biomed. Eng. 7, 1636–1648 (2023). Chen, S. et al. Present status of microfluidic PCR chip in nucleic acid detection and future perspective. Trac-Trend. Anal. Chem. 157, 116737 (2022). Gao, D. et al. Microfluidic chip and isothermal amplification technologies for the detection of pathogenic nucleic acid. J. Biol. Eng. 16, 33 (2022). Swanson, K. et al. From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment. Cell. 186, 1772–1791 (2023). Zhang, Z. et al. Machine Learning-Aided Identification of Fecal Extracellular Vesicle microRNA Signatures for Noninvasive Detection of Colorectal Cancer. ACS. Nano. 19, 10013–10025 (2025). Simmel, F. C. et al. Principles and Applications of Nucleic Acid Strand Displacement Reactions. Chem. Rev. 119, 6326–6369 (2019). Additional Declarations There is NO Competing Interest. Supplementary Files circRNAdetectionplatformSI20250418.docx Cite Share Download PDF Status: Posted Version 1 posted 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-6504183","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":446843824,"identity":"d4797a53-498a-4457-88e6-c1dd31841f1a","order_by":0,"name":"Lei Mou","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0ElEQVRIiWNgGAWjYBACPgbGBiBlw8AgAeYzE9bCBtGSRpIWMDhMihb25tYNH9vO282f3Z0mwVBhndjAfvYAfi08B9tuzmy7ndw45+w2CYYz6YkNPHkJ+LVIJLbd5gVqYZbI3SbB2HY4sUGCxwC/FvmHIC3nktnAWv4RowVoMlDLATsesJYGYrTwJLbdnHEuOUFCInezRcKxdOM2nhz8WvjZjz+78aHMzl5+Ru7GGx9qrGX72c/g1wIDiQ0gMoEBHlOEgT2xCkfBKBgFo2AEAgA520Ie7O6lGwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0003-4601-6991","institution":"Guangzhou Medical University","correspondingAuthor":true,"prefix":"","firstName":"Lei","middleName":"","lastName":"Mou","suffix":""},{"id":446843825,"identity":"bd7baaf8-87d2-44ad-a4ed-1665f26cb0ff","order_by":1,"name":"Xinyu Zhang","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xinyu","middleName":"","lastName":"Zhang","suffix":""},{"id":446843826,"identity":"c266dc95-a72d-4eee-b16c-4553fca2a630","order_by":2,"name":"Zixin Lin","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zixin","middleName":"","lastName":"Lin","suffix":""},{"id":446843827,"identity":"fb4a4f9a-ab80-41d7-9686-8ec7bdbf2706","order_by":3,"name":"Yongmei Chen","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yongmei","middleName":"","lastName":"Chen","suffix":""},{"id":446843828,"identity":"f16ba2c0-566e-4c35-81eb-4e5f4e54180c","order_by":4,"name":"Huanyu Zhou","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Huanyu","middleName":"","lastName":"Zhou","suffix":""},{"id":446843829,"identity":"41f63e96-f177-4b75-90fb-37fb05b3eb8d","order_by":5,"name":"yuetao zhang","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"yuetao","middleName":"","lastName":"zhang","suffix":""},{"id":446843830,"identity":"c2154692-8e77-4e6e-9b73-ea5d17bd80b5","order_by":6,"name":"Zijie Li","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zijie","middleName":"","lastName":"Li","suffix":""},{"id":446843831,"identity":"650e6f05-1315-420b-8e8b-d111a8f89222","order_by":7,"name":"Runlin Lv","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Runlin","middleName":"","lastName":"Lv","suffix":""},{"id":446843832,"identity":"3130b8f7-11f8-41c8-9e52-b4e1ac336901","order_by":8,"name":"Zhiqi Li","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zhiqi","middleName":"","lastName":"Li","suffix":""},{"id":446843833,"identity":"89060af1-3795-4cb9-bbda-ef2e02428693","order_by":9,"name":"Rubing Xiong","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Rubing","middleName":"","lastName":"Xiong","suffix":""},{"id":446843834,"identity":"d793e663-40fa-4bb9-ada1-a17a85f2a9dd","order_by":10,"name":"Yong Xia","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yong","middleName":"","lastName":"Xia","suffix":""},{"id":446843835,"identity":"dfce7d2e-89c4-4331-8ba1-b5fd46960c38","order_by":11,"name":"Honghai Hong","email":"","orcid":"","institution":"Guangzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Honghai","middleName":"","lastName":"Hong","suffix":""}],"badges":[],"createdAt":"2025-04-22 12:20:44","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6504183/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6504183/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82149430,"identity":"04404cbe-9992-4fe0-a44a-5de86a827f0f","added_by":"auto","created_at":"2025-05-07 07:09:28","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":105981,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFrom biomarker discovery to a micro-platform for BC diagnosis. a,\u003c/strong\u003e Schematic illustration of the biomarker discovery process for BC diagnosis. \u003cstrong\u003eb,\u003c/strong\u003e Electrochemical detection process based on CHA. \u003cstrong\u003ec,\u003c/strong\u003e ECMCDP detection procedures in blood and tissue samples of BC.\u003c/p\u003e","description":"","filename":"OnlineFig.1.png","url":"https://assets-eu.researchsquare.com/files/rs-6504183/v1/3234d3188f1e7ef2a8725178.png"},{"id":82149433,"identity":"73a1fe06-c6e4-472a-afd4-939a99b86fd3","added_by":"auto","created_at":"2025-05-07 07:09:28","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":131720,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDiscovery of novel biomarkers for BC diagnosis.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003e Heat map of differentially expressed circRNAs in the serum of BC patients and non-cancer healthy control (P \u0026lt; 0.05, FC \u0026gt; 1.5). \u003cstrong\u003eb, c,\u003c/strong\u003e qRT-PCR results of three up-regulated circRNAs (b) and three down-regulated circRNAs (c) verified the gene sequencing results. \u003cstrong\u003ed,\u003c/strong\u003e qRT-PCR results of has_circ_044235 (n = 5). \u003cstrong\u003ee,\u003c/strong\u003e ddPCR results of has_circ_044235 (n = 50). \u003cstrong\u003ef,\u003c/strong\u003eqRT-PCR results of has_circ_000250 (n = 30). \u003cstrong\u003eg,\u003c/strong\u003e ROC curves of has_circ_044235 and has_circ_000250. \u003cstrong\u003eh,\u003c/strong\u003e Anti-enzyme stability verification of hsa_circ_044235. \u003cstrong\u003ei,\u003c/strong\u003e Time stability verification of hsa_circ_044235. Error bars were defined by s.d. (mean ± s.d.).\u003c/p\u003e","description":"","filename":"OnlineFig.2.png","url":"https://assets-eu.researchsquare.com/files/rs-6504183/v1/01bd36179bdd9fa3f9dea6b4.png"},{"id":82149431,"identity":"c5b7954e-bf36-499d-91a4-c34e49a97ef6","added_by":"auto","created_at":"2025-05-07 07:09:28","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":319443,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCharacterization and structural analysis of AuPt-SPEs. a,\u003c/strong\u003e AuPt electrochemical \u003cem\u003ein situ\u003c/em\u003e deposition process. \u003cstrong\u003eb,\u003c/strong\u003e SEM images of bare SPEs.\u003cstrong\u003e c,\u003c/strong\u003e SEM image of AuPt-SPEs. \u003cstrong\u003ed,\u003c/strong\u003eAuPt-SPEs TEM images and characteristic lattice analysis of Au and Pt. \u003cstrong\u003ee,\u003c/strong\u003eSize distribution of AuPt nanoparticles measured by TEM (n = 85). \u003cstrong\u003ef,\u003c/strong\u003e AuPt-SPEs EDX layer image. \u003cstrong\u003eg,\u003c/strong\u003e EDS element image data corresponding to the AuPt-SPEs. \u003cstrong\u003eh, i,\u003c/strong\u003e AFM images of bare SPEs (h) and AuPt-SPEs (i).\u003c/p\u003e","description":"","filename":"OnlineFig.3.png","url":"https://assets-eu.researchsquare.com/files/rs-6504183/v1/83f18431cae73d65399df41c.png"},{"id":82149434,"identity":"bec7a61c-5f72-45d3-8875-b7effb8974fa","added_by":"auto","created_at":"2025-05-07 07:09:28","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":149950,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFeasibility of in vitro electrochemical biosensors.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003e Schematic diagram of the electrochemical biosensors construction process. \u003cstrong\u003eb, c,\u003c/strong\u003e The assembly of hsa_circ-044235-CHA (b) and hsa_circ-000250-CHA (c) was detected by Native-PAGE. \u003cstrong\u003ed, e,\u003c/strong\u003e Characterization of the working electrode for CV (d) in and OCP-EIS (e) in a 0.5 M KCl solution containing 10 mM K\u003csub\u003e4\u003c/sub\u003eFe(CN)\u003csub\u003e6\u003c/sub\u003e/K\u003csub\u003e3\u003c/sub\u003eFe(CN)\u003csub\u003e6\u003c/sub\u003e (1:1). \u003cstrong\u003ef, g,\u003c/strong\u003e DPV response curves of hsa_circ_044235 (f) and hsa_circ_000250 (g) at different concentrations in 10 mM Tris-HCl (pH = 7.4). \u003cstrong\u003eh, i,\u003c/strong\u003e Calibration curve of peak current versus lg C\u003csub\u003ehsa_circ_044235 \u003c/sub\u003e(h)\u003csub\u003e \u003c/sub\u003eand C\u003csub\u003ehsa_circ_000250 \u003c/sub\u003e(i). \u003cstrong\u003ej,\u003c/strong\u003e Storage stability of biosensors (n = 3). \u003cstrong\u003ek,\u003c/strong\u003e Reproducibility of the biosensors (n = 3). \u003cstrong\u003el, n,\u003c/strong\u003e Selectivity of electrochemical biosensors for hsa_circ_044235 (l) and hsa_circ_000250 (n) with a concentration of 100 nM for circRNAs and 1 μM for other interfering substances (n = 3). Error bars were defined by s.d. (mean ± s.d.).\u003c/p\u003e","description":"","filename":"OnlineFig.4.png","url":"https://assets-eu.researchsquare.com/files/rs-6504183/v1/e1088ab942768975c79142b6.png"},{"id":82150603,"identity":"b5a39cda-0390-4d5d-a5fc-d734c4ef599f","added_by":"auto","created_at":"2025-05-07 07:17:28","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1651004,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDesign and manufacture of ECMCDP. a,\u003c/strong\u003eComposition of ECMCDP. \u003cstrong\u003eb,\u003c/strong\u003e The layering schematic of the microfluidic chip. \u003cstrong\u003ec,\u003c/strong\u003e Images of the concentration (i) and velocity (ii) fields simulated by COMSOL software. \u003cstrong\u003ed,\u003c/strong\u003e Physical image of the reaction in a microfluidic chip. \u003cstrong\u003ee,\u003c/strong\u003e The system block diagram of FPCB. \u003cstrong\u003ef, \u003c/strong\u003eComparison of detection results between ECMCDP and commercial electrochemical workstations (n = 3). Error bars were defined by s.d. (mean ± s.d.).\u003c/p\u003e","description":"","filename":"OnlineFig.5.png","url":"https://assets-eu.researchsquare.com/files/rs-6504183/v1/ec51cecd4439e5c36d4d5235.png"},{"id":82149432,"identity":"e031feeb-0702-437f-b1e6-11dd243861e9","added_by":"auto","created_at":"2025-05-07 07:09:28","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":99722,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClinical feasibility of ECMCDP.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003e Schematic representation of clinical samples collected in hospitals tested side-by-side with qRT-PCR and ECMCDP. \u003cstrong\u003eb, c,\u003c/strong\u003e Results of hsa_circ_044235 (b) and hsa_circ_000250 detected by ECMCDP in the blood of BC patients and HCs (BC = 44, HCs = 36). \u003cstrong\u003ed,\u003c/strong\u003eSensitivity and specificity of ROC curve for the diagnosis of ECMCDP (n = 40). \u003cstrong\u003ee,\u003c/strong\u003eConsistency of clinical diagnostic results of ECMCDP results. \u003cstrong\u003ef, g,\u003c/strong\u003e The results of hsa_circ_044235 (f) and hsa_circ_000250 (g) in the blood of BC patients before and after the operation were detected by ECMCDP (n = 4). \u003cstrong\u003eh, i,\u003c/strong\u003eECMCDP results of hsa_circ_044235 (h) and hsa_circ_000250 (i) in normal tissues, pare-carcinoma tissues, and carcinoma tissues of BC patients (n = 7).\u003c/p\u003e","description":"","filename":"OnlineFig.6.png","url":"https://assets-eu.researchsquare.com/files/rs-6504183/v1/fa54e83be3a83afefd84c89d.png"},{"id":82149441,"identity":"bf4ebca0-7982-4172-b9c8-488aad879625","added_by":"auto","created_at":"2025-05-07 07:09:28","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":1448399,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eML for BC diagnosis.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003e Schematic diagram of the research ML process. \u003cstrong\u003eb,\u003c/strong\u003e The association between biomarkers and subjects (patients) in the chord diagram. \u003cstrong\u003ec,\u003c/strong\u003e Principal component analysis (PCA) scatter plot based on the two markers (red represents BCs and blue represents HCs). \u003cstrong\u003ed,\u003c/strong\u003e Heat map of correlation between different patients based on all biochemical index data. \u003cstrong\u003ee,\u003c/strong\u003e Pearson correlation heatmap between biomarkers. \u003cstrong\u003ef,\u003c/strong\u003e Results of feature importance analysis of the model. \u003cstrong\u003eg,\u003c/strong\u003e TP, FP, TN, and FN on the validation set (20 % data) using the four analytical ensemble models. \u003cstrong\u003eh,\u003c/strong\u003e The classification accuracy of hsa_circ_044235 and hsa_circ_000250 for BC with five different models. \u003cstrong\u003ei,\u003c/strong\u003e Confusion matrix of the ensemble model on the validation set. \u003cstrong\u003ej,\u003c/strong\u003e ROC curves comparison between deep neural network model and soft voting ensemble model on the validation set.\u003c/p\u003e","description":"","filename":"OnlineFig.7.png","url":"https://assets-eu.researchsquare.com/files/rs-6504183/v1/17ec24a2c8e3c6c00e0275e7.png"},{"id":82659197,"identity":"90e54414-fa6f-4aac-93e4-9e99b883c001","added_by":"auto","created_at":"2025-05-13 20:03:56","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3036671,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6504183/v1/15ef3377-d26c-44ac-acf0-478addca4659.pdf"},{"id":82150607,"identity":"43323548-80fb-4867-aeef-ef3107cc6a66","added_by":"auto","created_at":"2025-05-07 07:17:28","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":37090819,"visible":true,"origin":"","legend":"","description":"","filename":"circRNAdetectionplatformSI20250418.docx","url":"https://assets-eu.researchsquare.com/files/rs-6504183/v1/8f03a5179fe9fc00debb73c5.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"A Novel Machine Learning-enhanced Microfluidic CircRNAs Detection Platform for Breast Cancer Precision Diagnosis","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eBreast cancer (BC) is the most commonly diagnosed cancer and the leading cause of cancer death among women in 2022\u003csup\u003e1,2\u003c/sup\u003e. With the in-depth study of the molecular characteristics of BC, the diagnosis and classification methods of BC have changed significantly. New biomarkers are essential to improve diagnostic accuracy, early detection, and personalized treatment\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e,\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Circular RNA (circRNA) are exonuclease-resistant, covalently closed RNAs generated via backsplicing, lacking terminal structures (5\u0026prime; caps/3\u0026prime; poly-A tails). Their stability and ubiquitous presence in bodily fluids highlight their potential as cancer biomarkers for diagnostic and therapeutic monitoring \u003csup\u003e\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Therefore, the discovery of circRNAs closely related to BC and the realization of its highly sensitive and specific detection are essential for the early diagnosis of BC. However, the existing detection methods have limitations such as cumbersome operation, long time, high background signal, and complex experiment, which hinder the rapid clinical detection application\u003csup\u003e\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. It is urgent to develop simple, economical, high-sensitivity, and high-specificity detection methods.\u003c/p\u003e \u003cp\u003eIsothermal nucleic acid amplification is a promising nucleic acid detection method that can detect DNA, RNA, cells, and other biological targets. It has the advantages of rapid, low cost, simple operation and is suitable for field detection\u003csup\u003e\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. The technique includes enzyme-assisted amplification and enzyme-free amplification. Enzyme-assisted amplification methods such as rolling ring amplification (RCA) and loop-mediated isothermal amplification (LAMP) have disadvantages such as high cost, difficult storage, and complex reaction conditions\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Compared with other enzyme-free amplification methods, including hybrid chain reaction (HCR), DNAzyme, and entropy-driven circuit (EDC), catalytic hairpin assembly (CHA) has a simpler and more stable reaction system and higher catalytic efficiency\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eElectrochemical biosensors combine electronic transduction systems with biorecognition elements, achieving sensitive signal detection through electrochemical methods and molecular specificity via selective bioaffinity interactions\u003csup\u003e\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. Surface amplification of nucleic acids enables the integration of electrochemical biosensors and targeted amplification methods on the same platform. This integration reduces assay time, reduces the risk of contamination, and also allows point-of-care (POC) molecular detection\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. The electrochemical conversion strategies are classified into labeling-free (studying the rate of electron transfer between redox pairs and modified electrode surfaces in solution) and labeling-based (using redox substances such as methylene blue (MB) and ferrocene (Fc)) methods. These substances bind to the generated long single-stranded DNA (ssDNA)\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. ssDNA carries many negative charges, and the sequence can be designed to be an excellent biopolymer. Whose assembly is driven by metal ion-DNA coordination and electrostatic force\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Various noble metal nanoparticles, like gold nanoparticles (AuNPs) and platinum nanoparticles (PtNPs), improve current response due to their conductivity and catalytic properties\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e,\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. Moreover, AuNPs and PtNPs can bind to thiol-modified ssDNA\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. Pt-S bonds are more stable than Au-S bonds. Adding Pt to AuNPs forms stable bioconjugates and reduces interference from biothiols\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e,\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. Compared with single-metal nanoparticles, bimetallic nanoparticles are more stable and biocompatible with specific optical or electrical activities\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn this study, we discovered two BC-related circRNAs and established a new highly sensitive circRNAs detection platform. Circular RNA microarray technology was used to analyze circRNAs expression profiles in BC serum. The value of hsa_circ_044235 and hsa_circ_000250 in the diagnosis of BC was verified by digital PCR (ddPCR) and real-time fluorescent quantitative reverse transcription polymerase chain reaction (qRT-PCR) (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). Traditional qRT-PCR is complex and subject to contamination risks, microfluidic integrated isothermal amplification technology provides a portable, automated solution for POC detection and cancer screening\u003csup\u003e\u003cspan additionalcitationids=\"CR30 CR31\" citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. Further, we developed a highly sensitive and specific electrochemical microfluidic circRNA detection platform, used gold platinum nanoparticles (AuPts)-modified screen-printed electrodes (SPEs), catalytic hairpin assembly (CHA), and a custom-designed low-power electronics system (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb). It was successfully applied to the detection of clinical blood and tissue samples (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec). Machine Learning (ML) has significantly improved the accuracy and efficiency of tumour diagnosis, enabling early detection and personalized treatment by analyzing medical imaging and molecular data, and promoting multimodal data integration and clinical translation\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e,\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. This study used five machine learning models to achieve efficient, accurate, and economical breast cancer detection and prediction, and the accuracy of the ensemble model was 93.75%. Additionally, it offers a theoretical and practical basis for identifying other disease biomarkers.\u003c/p\u003e"},{"header":"2. Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Discovery of new biomarkers for BC diagnosis\u003c/h2\u003e \u003cp\u003eTo explore the difference in serum circRNAs expression between the healthy control group (HCs) and BC patients group. We randomly collected serum samples from 3 HCs without cancer and 3 patients with BC. Arraystar human circRNAs microarray technology was used to sequence 13,433 circRNAs in serum, of which 7329 were up-regulated, and 614 were down-regulated. According to P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, fold change (FC)\u0026thinsp;\u0026gt;\u0026thinsp;1.5, a total of 105 differentially expressed circRNAs were identified, including 54 up-regulated circRNAs and 51 down-regulated circRNAs. Significantly differentially expressed circRNAs were selected (\u003cb\u003eSupplementary Fig.\u0026nbsp;1a, Supplementary Table\u0026nbsp;1\u003c/b\u003e). The heat map showed the top 50 differentially expressed circRNAs (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea).\u003c/p\u003e \u003cp\u003eTo verify the accuracy of circRNAs preliminary screening results, we selected three up-regulated circRNAs (hsa_circ_102101, hsa_circ_104293, hsa_circ_400241) and down-regulated circRNAs (hsa_circ_008053, hsa_circ_008016, hsa_circ_103637) were verified by qRT-PCR. The qRT-PCR results were consistent with the gene sequencing results (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb, c). The results showed that the sequencing results were accurate and reliable. Based on its most significant differential expression profile in serum (FC\u0026thinsp;\u0026gt;\u0026thinsp;2.99, P\u0026thinsp;\u0026lt;\u0026thinsp;0.05), hsa_circ_044235 was prioritized for subsequent studies. Sequencing results showed that hsa_circ_000250 was also differentially expressed in BC patients (\u003cb\u003eSupplementary Fig.\u0026nbsp;1b\u003c/b\u003e). The qRT-PCR results showed that hsa_circ_044235 expression decreased in serum of BC patients compared with HCs (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ed). Meanwhile, we constructed a ddPCR method for the detection of hsa_circ_044235. Explored the optimal annealing temperature and total sample size and selected 42.8\u0026deg;C as the optimal annealing temperature (\u003cb\u003eSupplementary Fig.\u0026nbsp;2a, b\u003c/b\u003e) and 5 ng as the optimal total sample size (\u003cb\u003eSupplementary Fig.\u0026nbsp;2c, d\u003c/b\u003e). ddPCR showed decreased expression of hsa_circ_044235 in the serum of BC patients compared with HCs (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ee). qRT-PCR results showed that hsa_circ_000250 was increased in the serum of BC patients compared with HCs (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ef). For all HCs and BC patients, the area under the curve (AUC) values of hsa_circ_044235 and hsa_circ_000250 were 0.87 (95% confidence interval (CI) 0.83\u0026ndash;0.91) and 0.88 (95% CI 0.81\u0026ndash;0.95) respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eg). Furthermore, compared with linear RNA and human-β-actin, circRNA exhibited higher temporal stability and better resistance to enzymatic degradation (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eh, i). These results indicated that hsa_circ_044235 and hsa_circ_000250 show significant expression differences in BC patients, and combined with their high diagnostic sensitivity and specificity, they can be used as effective biomarkers for breast cancer diagnosis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Preparation and characterization of electrochemical biosensors\u003c/h2\u003e \u003cp\u003eThe electrochemical biosensor was fabricated using a multilayer screen printing method, with conductive ink, Ag/AgCl reference ink, and UV insulating ink printed on polyethylene glycol terephthalate (PET) film, respectively (\u003cb\u003eSupplementary Fig.\u0026nbsp;3\u003c/b\u003e). Before the sensor was functionalized, we activated the SPEs using the oxygen plasma method (\u003cb\u003eSupplementary Fig.\u0026nbsp;4a\u003c/b\u003e). Explored the plasma treatment time, the SPEs surface current value achieved maximum when plasma treatment time was 6 min (\u003cb\u003eSupplementary Fig.\u0026nbsp;4b, c\u003c/b\u003e). The electrical conductivity of the SPEs was significantly increased by plasma activation (\u003cb\u003eSupplementary Fig.\u0026nbsp;4d\u003c/b\u003e). SPEs were hydrophobic with a contact Angle (CA) of 110 (\u003cb\u003eSupplementary Fig.\u0026nbsp;5a\u003c/b\u003e). After plasma, the SPEs exhibited a cleaner surface with increased hydrophilicity (CA\u0026thinsp;=\u0026thinsp;39.52) (\u003cb\u003eSupplementary Fig.\u0026nbsp;5b\u003c/b\u003e).AuNPs and PtNPs were deposited \u003cem\u003ein situ\u003c/em\u003e on the SPEs by cyclic voltammetry (CV) method to increase the electrochemical performance of the SPEs (\u003cb\u003eSupplementary Fig.\u0026nbsp;6\u003c/b\u003e). Moreover, the concentration, deposition cycles and ratio of Au:Pt were optimized. The current value of the SPEs surface achieved the maximum under the optimized conditions. (The concentrations of HAuCl\u003csub\u003e4\u003c/sub\u003e\u0026middot;3H\u003csub\u003e2\u003c/sub\u003eO and K\u003csub\u003e2\u003c/sub\u003ePtCl\u003csub\u003e4\u003c/sub\u003e were 1 mM and 5 mM, respectively, (\u003cb\u003eSupplementary Fig.\u0026nbsp;7\u003c/b\u003e), with 15 deposition cycles (\u003cb\u003eSupplementary Fig.\u0026nbsp;8\u003c/b\u003e) and an Au:Pt ratio of 1:1 (\u003cb\u003eSupplementary Fig.\u0026nbsp;9a\u003c/b\u003e)). Meanwhile, the biosensor response to hsa_circ_000250 was maximum when the Au:Pt ratio was 1:1 (\u003cb\u003eSupplementary Fig.\u0026nbsp;9b\u003c/b\u003e). The CA on the SPEs surface after AuPt deposition was 68.20 (\u003cb\u003eSupplementary Fig.\u0026nbsp;5c\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eAuPt was deposited on SPEs \u003cem\u003ein situ\u003c/em\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea). The morphology and chemical composition of AuPt-SPEs were characterized. Compared with bare SPEs, SEM showed that AuPt nanoparticles were successfully and evenly distributed on the SPE surface (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb, c). The successful fusion of AuNPs with PtNPs nanoparticles were confirmed by high-resolution TEM imaging, which showed lattice stripes in different states of Au and Pt (\u003cb\u003eSupplementary Fig.\u0026nbsp;10\u003c/b\u003e, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ed). The diameter distribution of AuPt nanoparticles ranged from 20 to 80 nm (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ee). EDX and EDS analysis further demonstrated the successful fusion of AuNPs and PtNPs (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ef, g). AFM was used to characterize the average surface roughness of bare SPEs and AuPt-SPEs. Compared with the bare SPE, the surface roughness of AuPt-SPEs is significantly increased (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eh, i). After AuPt was confirmed to be successfully deposited on the SPE surface. CV was used to test the electrochemical surface active area of the bare SPEs and the AuPt-SPEs, which were 0.006 cm\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e and 0.037 cm\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e, respectively. Compared with the bare SPEs, the electrochemical surface active area of AuPt-SPEs increased by nearly 5 times. It is further indicated that the deposition of AuPt on the SPEs surface significantly increases the electrochemical active sites and improves the electrochemical performance of the SPEs.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Feasibility of electrochemical biosensors\u003c/h2\u003e \u003cp\u003eThe design of the electrochemical biosensor is based on attaching a thiolated hairpin probe (SH-H1/H1') to the AuPt-SPEs surface. The target circRNA can open and bind to the H1/H1' hairpin structure. Another ferrocene-labeled hairpin probe (Fc-H2/H2') can form a double-stranded structure with the open H1/H1' and gradually replace the release target circRNA. The Fc-H2/H2' is close to the AuPt-SPEs. The detected electrochemical signal is proportional to the target circRNA content (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). The feasibility of our electrochemical biosensor was initially demonstrated using native polyacrylamide gel electrophoresis (Native-PAGE). Only one band was visible when H1/H1', H2/H2', or circRNA was present alone. There are two separate bands in the presence of both H1/H1' and H2/H2', proving that no hybridization between H1 and H2 occurs without the target circRNA. A new band with slow electrophoretic mobility and large molecular weight appeared after the addition of the target circRNA, indicating that the addition of the target circRNA triggered the CHA process of H1 and H2 (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb, c). Further, the ΔG of H1-H2, H1'-H2' combined is smaller than that of H1/H1', H2/H2', and H1/H1'-circRNA, demonstrating the feasibility of electrochemical biosensors\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e(\u003cb\u003eSupplementary Fig.\u0026nbsp;11, 12\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eCV and open circuit potential-electrochemical impedance spectroscopy (OCP-EIS) were used to further characterize the surface of AuPt-SPEs after each modification step (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed, e). When AuPt original was deposited on the surface of SPEs, the peak current height increased in the CV diagram, and the resistance decreased in the Nyquist diagram. These changes indicated that the conductivity and electrochemical catalytic activity of the SPEs increased. Subsequent modifications on the surface of the AuPt-SPEs prevent charge transfer on the electrode surface, resulting in lower peak current height in the CV diagram and higher resistance in the Nyquist diagram. After confirming the successful modification of AuPt-SPEs, the probe concentration, reaction time, and reaction temperature were further optimized to achieve the best detection performance (\u003cb\u003eSupplementary Fig.\u0026nbsp;13, 14\u003c/b\u003e). Both circRNAs showed the best detection performance under the condition that the probe concentration was 2 \u0026micro;M and the reaction time was 37\u0026deg;C for 2 h. The electrochemical biosensor performance was evaluated by differential pulse voltammetry (DPV) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ef, g). The electrochemical biosensor showed a log-linear relationship between the peak current height of DPV and the target concentration. The ultra-low LOD of hsa_circ_004423 and hsa_circ_000250 detected by electrochemical biosensors were 0.12 fM and 0.01 fM, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eh, i). Stability was assessed by testing the DPV response of AuPt-SPEs placed at 4\u0026deg;C from 1 to 14 days, with a relative standard deviation (RSD) of 4.44% at 14 days (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ej). Five SPEs were prepared in parallel to detect different concentrations of circRNA (100 nM, 10 nM) to evaluate the reproducibility of the electrochemical biosensor, with RSD of 3.91% and 5.52% for the two concentrations, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ek). The electrochemical biosensor showed high selectivity in the presence of high concentrations of interfering substances (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003el, n). These results verified the feasibility of the reaction system and proved that the biosensor has remarkable stability, reproducibility and specificity. Demonstrated its ability to meet actual testing requirements.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Design and fabrication of ECMCDP\u003c/h2\u003e \u003cp\u003eTo realize low-cost, simple and integrated portable detection, this study designed a compact wireless device integrating a microfluidic chip and a flexible printed circuit board (FPCB). The results were compared with those of a commercial electrochemical workstation (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea, \u003cb\u003eSupplementary Fig.\u0026nbsp;15a)\u003c/b\u003e. The microfluidic chip consists of a microchannel and microchamber layer, an electrode layer, VHB glue, and a PDMS support layer (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb). The microfluidic chip comprised a 10 \u0026micro;L H1/H1' prestorage chamber, a 10 \u0026micro;L H2/H2' prestorage chamber, a 20 \u0026micro;L sample RNA reservoir, a mixing channel, an electrochemical reaction zone, a 100 \u0026micro;L wash chamber, and a 200 \u0026micro;L waste collection chamber. Under the action of a micro-peristaltic pump, the reagents are transported from different storage areas through the mixing channel to the reaction area, triggering the CHA reaction with H1/H1' fixed on the AuPt-SPEs. (\u003cb\u003eSupplementary Fig.\u0026nbsp;15b\u003c/b\u003e). Using COMSOL software to simulate the mixing effect of mixing channels. In the designed mixing channel, two liquids of different concentrations mix well and reach the reaction zone at a constant speed (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec). Since the CHA reaction required 2 hours at 37\u0026deg;C, the stable liquid volume in the microfluidic system confirmed effective sealing and compatibility with optimal reaction conditions. (\u003cb\u003eSupplementary Fig.\u0026nbsp;16\u003c/b\u003e). Red and blue inks simulated the reaction process, with reagents preloaded in storage chambers. After the micro peristaltic pump was activated, liquids flowed through the mixing channel into the reaction zone. Post-reaction, a washing solution was introduced to quench the reaction (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ed). The microfluidic system was connected to a reusable PCB potentiostat system. The PCB potentiostat system block diagram (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ee) and circuit diagram (\u003cb\u003eSupplementary Fig.\u0026nbsp;17\u003c/b\u003e) were used to measure the electrical response signal during the process. In the MCU-based FPCB control system, the voltage output is programmed by the integrated DAC, and the potentiostat module and electrode selection module drive the sensor electrodes. The current signal of the impedance amplifier (precision operational amplifier with analog switches) was collected by the ADC and transmitted to the smartphone via Bluetooth for real-time monitoring and analysis. The DPV results of ECMCDP were compared with those of commercial electrochemical workstations. These results indicated the reliability of the output results of ECMCDP (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ef). The microfluidic system costs only \u003cspan\u003e$\u003c/span\u003e2.08, and the FPCB can be reused (\u003cb\u003eSupplementary Table\u0026nbsp;2\u003c/b\u003e). It can provide an economical and feasible technical scheme for large-scale BC screening programs.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Clinical feasibility of ECMCDP\u003c/h2\u003e \u003cp\u003eTo preliminarily assess the potential clinical application of ECMCDP in BC diagnosis, blood samples were collected from 44 BC patients and 36 HCs. Additionally, we obtained surgically - removed normal, para-carcinoma, and carcinoma tissues from 7 BC patients. The results of ECMCDP were compared with those of traditional qRT-PCR. Compared with traditional qRT-PCR, which requires a complex operation process such as reverse transcription, ECMCDP can directly detect total RNA and output the results (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea). ECMCDP successfully detected hsa_circ_044235 and hsa_circ_000250 in the blood of HCs and BC patients. The expression trends of the two circRNAs were consistent with our previous verification results (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb, c). Meanwhile, correlation analysis was conducted between the current values of the two circRNAs and Ct values (\u003cb\u003eSupplementary Table\u0026nbsp;3\u003c/b\u003e). It can be seen from the figure that the changes between the current values of the two cricRNAs and Ct values were consistent and showed a significant negative correlation. Spearman's r values were \u0026minus;\u0026thinsp;0.6256 and \u0026minus;\u0026thinsp;0.5942, respectively \u003cb\u003e(Supplementary Fig.\u0026nbsp;18)\u003c/b\u003e. This indicates that the higher the Ct value, the lower the circRNAs expression level and the corresponding current value. Moreover, ECMCDP, even at high Ct values (\u0026gt;\u0026thinsp;30), can also detect the current signal.\u003c/p\u003e \u003cp\u003eROC curves of ECMCDP detection were plotted, and the AUC were calculated to discuss the diagnostic accuracy of the ECMCDP (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ed, \u003cb\u003eSupplementary Table\u0026nbsp;4\u003c/b\u003e). The AUC values of hsa_circ_044235 and hsa_circ_000250 were 0.98 (95% CI 0.94\u0026ndash;1.00) and 0.92 (95% CI 0.81\u0026ndash;1.00), respectively. Compared with other commonly used clinical protein biomarkers, the AUC of CEA, CA125, and CA153 were respectively 0.86 (95% CI 0.74\u0026ndash;0.98), 0.82 (95% CI 0.70\u0026ndash;0.95) and 0.67 (95% CI 0.50\u0026ndash;0.85). Results from the study samples of HCs and BC patients demonstrate the potential of these two circRNAs as diagnostic biomarkers for BC. They also showed the practical application of the ECMCDP for the early diagnosis of BC. The AUC of the combined detection of two circRNAs and other protein biomarkers was 1.00, indicating that the collaborative detection between biomarkers can significantly improve the diagnostic efficiency of BC \u003cb\u003e(Supplementary Table\u0026nbsp;5)\u003c/b\u003e. The ROC curves provided cut-off current values of 0.17 and 0.19 for the sum of optimized sensitivity and specificity. Using these cut-off values, ECMCDP can accurately distinguish positive and negative clinical samples with an accuracy of 92.50% and 88.75%, respectively, with high sensitivity and specificity (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ee, \u003cb\u003eSupplementary Table\u0026nbsp;6\u003c/b\u003e). ECMCDP stands out for its high sensitivity and accuracy compared to other methods for detecting breast cancer-related biomarkers (\u003cb\u003eSupplementary Table\u0026nbsp;7\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eMonitoring biomarker levels after treatment can indicate response evaluation, predict disease prognosis, and guide subsequent individualized therapy. Therefore, we collected preoperative and postoperative blood samples from 4 BC patients and used ECMCDP to detect the changes of two circRNAs before and after the operation. Compared with the preoperative results, the expression of hsa_circ_044235 in the blood of BC patients increased, and hsa_circ_000250 decreased, indicating their potential as effective monitoring biomarkers of BC (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ef, g, \u003cb\u003eSupplementary Table\u0026nbsp;8\u003c/b\u003e). ECMCDP was also used to detect the expression of two circRNAs in normal tissues, para-carcinoma tissues, and carcinoma tissues of 7 BC patients. Compared with normal and para-carcinoma tissues, hsa_circ_044235 was up-regulated, and hsa_circ_000250 was down-regulated in carcinoma tissues. Contrary to the trend of expression in blood, this suggests the possibility of both circRNAs as BC tissue-specific biomarkers (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eh, i, \u003cb\u003eSupplement Table\u0026nbsp;9a, b\u003c/b\u003e). The ECMCDP demonstrated high specificity and sensitivity in detecting hsa_circ_004423 and hsa_circ_000250 across diverse human biological samples, indicating broad clinical utility.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 ML for BC diagnosis\u003c/h2\u003e \u003cp\u003eGradient boosting (GB), support vector machine (SVM), random forest (RF), and logistic regression (LR) models were used to process the current signals of hsa_circ_044235 and hsa_circ_000250 derived from ECMCDP. Soft voting was used to predict the diagnosis of BC (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea). The chord chart showed the association of biomarkers with patients (bandwidth\u0026thinsp;=\u0026thinsp;relative contribution). It can be seen that circRNA markers are mainly closely associated with BC patient samples, and hsa_circ_000250 is weaker than hsa_circ_044235 with patients (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb). Principal component analysis (PCA) showed that BC patients (n\u0026thinsp;=\u0026thinsp;44) and HCs (n\u0026thinsp;=\u0026thinsp;36) were clearly separated in 2D PCA space (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ec). Correlation heatmaps for all biochemical markers, including circRNAs and traditional biomarkers, revealed heterogeneity among patients and potential typing trends through Pearson correlation coefficients (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ed). The Pearson correlation heatmap between the biochemical markers showed the correlation coefficients of their expression change trends across all samples, highlighting potential synergistic or complementary relationships (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ee). Feature importance analysis of the model highlighted hsa_circ_044235 and hsa_circ_000250 as the most contributing features in the model (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ef). SVW, LR, RF, and GB were used to analyze the true positive (TP), false positive (FP), true negative (TN), and false negative (FN) (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eg) of the ensemble model on the validation set (20% data). Five ML models achieved BC classification accuracies of 86.67%, 87.50%, 93.75%, 100%, and 100%, with the ensemble model reached 93.75% accuracy (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eh). Confusion matrix analysis of the validation set showed high sensitivity and specificity for correct classification of patients and healthy individuals (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ei). Comparison of the ROC curves of the deep neural network model and the soft voting ensemble model on the validation set showed that the ensemble model had robust diagnostic performance with an AUC of 0.94 (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ej). Therefore, ML provides a robust and scalable computational framework to systematically evaluate the performance of enhanced ECMCDP in detecting circRNAs.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"3. Conclusion","content":"\u003cp\u003eIn this study, we identified two novel circRNAs (hsa_circ_044235 and hsa_circ_000250) as highly sensitive and specific serum biomarkers for BC diagnosis. Through comprehensive circRNA microarray screening and subsequent validation via qRT-PCR and ddPCR, we demonstrated that these circRNAs exhibit significant differential expression between BC patients and HCs, with high diagnostic accuracy (AUC\u0026thinsp;=\u0026thinsp;0.87, 0.88, respectively). Leveraging their stability and resistance to enzymatic degradation, we developed an ultrasensitive electrochemical biosensor based on CHA, achieving femtomolar-level detection limits (0.12 fM and 0.01 fM, respectively).\u003c/p\u003e \u003cp\u003eTo enable point-of-care testing, we designed an integrated, low-cost electrochemical microfluidic circRNA detection platform, which combines a microfluidic chip with a portable potentiostat system for automated, high-throughput analysis. Clinical validation in 44 BC patients and 36 HCs confirmed the platform\u0026rsquo;s diagnostic efficacy, with AUC values of 0.98 and 0.92 for hsa_circ_044235 and hsa_circ_000250, respectively-outperforming conventional protein biomarkers (CEA, CA125, CA153). Furthermore, this platform demonstrated excellent reproducibility, stability, clinical feasibility in monitoring postoperative treatment responses and distinguishing cancerous from non-cancerous tissues.\u003c/p\u003e \u003cp\u003eThe integration of ML enhanced diagnostic precision, with ensemble models (GB, SVM, RF, LR, Neural Network) achieving 93.75% accuracy in classifying BC patients. Feature importance analysis confirmed that the two circRNAs were the most predictive biomarkers, reinforcing their clinical utility. Additionally, this platform successfully monitored postoperative biomarker dynamics, suggesting its applicability in treatment response assessment.\u003c/p\u003e \u003cp\u003eThe discovery of circRNA biomarkers and their integration into an ML-powered microfluidic platform paves the way for large-scale, early-stage BC screening, with potential extensions to other cancers. Future studies should explore the mechanistic roles of these circRNAs in BC progression and validate this platform in larger, multi-center cohorts. This work represents a significant advancement in liquid biopsy-based cancer diagnostics, offering a transformative approach to precision oncology through biomarker discovery, biosensor engineering, and ML-driven data analysis.\u003c/p\u003e"},{"header":"4. Methods","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Materials and reagents\u003c/h2\u003e \u003cp\u003eAll DNA and circRNAs were custom-synthesized by Sangon Biotechnology Co., Ltd. (Shanghai, China) and dissolved in a DEPC-treated aqueous solution. The RNA solution can be used directly without treatment (\u003cb\u003eSupplementary Table\u0026nbsp;10\u003c/b\u003e). DNA solutions were formed into hairpin structures by heating at 95\u0026deg;C for 5 min followed by slow cooling at 6\u0026deg;C/min to room temperature. Tris (2-carboxyethyl) phosphine hydrochloride (TCEP) was purchased from MedChemExpress. 6-sulfhydryl-1-hexanol (MCH) and gold (III) chloride trihydrate (HAuCl\u003csub\u003e4\u003c/sub\u003e\u0026middot;3H\u003csub\u003e2\u003c/sub\u003eO) were purchased from Sigma-Aldrich. Potassium chlorite (K\u003csub\u003e2\u003c/sub\u003ePtCl\u003csub\u003e4\u003c/sub\u003e) was purchased from Aladdin. DNA-free ribonuclease water was purchased from Yitao Biotechnology Co., Ltd. (Guangzhou, China). Carbon ink (CH-8) and UV ink were purchased from China Pinghu Juju Consumable Technology (Pinghu) Co., Ltd. Ag/AgCl ink (011464) was purchased from ALS, Tokyo, Japan. The wash solution contained 10 mM Tris-Hcl, 50 mM NaCl, and 0.05% TWeen-20. RNA extraction kit, reverse transcription kit, and SYBR Green Ⅰ were purchased from Accurate Biology Co., Ltd. (Hunan, China). Digital PCR probe method supermix, probe method microdroplet generation oil, and microdroplet reading oil were purchased from Bio-rad, USA. The polydimethylsiloxane (PDMS; Sylgard 184) kit was purchased from Dow Corning.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Instruments\u003c/h2\u003e \u003cp\u003eCircRNAs in serum were sequenced using Arraystar human circRNAs microarray technology (China Kangcheng Biotechnology Co., Ltd.). ddPCR was performed using a Bio-Rad QX200 droplet digital PCR system, USA. qRT-PCR was performed using the QuantStudio 3 real-time PCR System. Native-PAGE was performed using the Bio-Rad Mini-PROTEAN Tetra System, USA.\u003c/p\u003e \u003cp\u003eElectrochemical characterization, including DPV, CV, OCP-EIS, was performed on a CHI650E electrochemical workstation (Shanghai Chenhua Instrument Co., Ltd., China). Plasma instruments were purchased from Zhongxinqinheng Co., Ltd. (Suzhou, China). The morphology of AuPt, as well as its dimensions, were characterized using a scanning electrode (Thermo Fisher TF20) and a transmission electron microscope (Zeiss Gemini SEM360). The surface roughness was analyzed by atomic force microscopy (NT-MDT, Russia). A miniature peristaltic pump (BW100-WP110) was purchased from Baoding Chuangrui Pump Co., Ltd.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Clinical blood and tissue samples collection\u003c/h2\u003e \u003cp\u003eThe blood and tissue samples of surgically resected BC patients diagnosed by pathology in the Third Affiliated Hospital of Guangzhou Medical University from September 2024 to March 2025 were collected. The blood samples of healthy people were collected from the Physical Examination Center of the Third Affiliated Hospital of Guangzhou Medical University. After collection and repacking, all samples were stored at -80\u0026deg;C. This study was approved by the Ethics Committee of the Third Affiliated Hospital of Guangzhou Medical University (2025048).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Serum circRNAs sequencing analysis\u003c/h2\u003e \u003cp\u003eIn order to explore the difference in the expression of circRNAs in the serum of HCs group and patient group. We randomly collected serum samples from 3 non-cancer healthy individuals and 3 BC patients. Arraystar human circRNAs microarray technology was used to sequence circRNAs in serum. The FC\u0026thinsp;\u0026gt;\u0026thinsp;1.5 and P\u0026thinsp;\u0026lt;\u0026thinsp;0.05 circRNAs were used as significantly differentially expressed circRNAs.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.5 Preliminary validation of clinical samples\u003c/h2\u003e \u003cp\u003eqRT-PCR verification of samples after sample collection, total RNA was extracted using an RNA extraction kit (AG21024). After reverse transcription (AG11728), 20 \u0026micro;L of the reaction system was prepared according to the qRT-PCR reagent instructions (AG11718). After mixing, the samples were detected using QuantStudio 3 real-time PCR system.\u003c/p\u003e \u003cp\u003eFor ddPCR verification, the samples were prepared into a 20 \u0026micro;L ddPCR reaction system on ice. Oil was generated by adding micro-droplets. The micro-droplets were covered with adhesive strips and placed in the micro-droplet generator to form water-in-oil micro-droplets. Four micro-droplets were transferred to a 96-well plate and sealed with a heat-sealing membrane. Amplification was performed using a Bio-Rad T100 gradient PCR apparatus. After amplification, the samples were detected and analyzed by a QX200 droplet digital PCR instrument.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.6 Preparation of AuPt-SPEs\u003c/h2\u003e \u003cp\u003eSPEs were designed using Auto CAD 2024 software. hsa_circ _044235 and hsa_circ _000250 shared a reference electrode and counter electrode, and conductive ink CH-8 was printed on the PET film to form the electrode circuit. After printing, it was oven-dried at 80\u0026deg;C for 20 min, followed by screen printing using UV insulating ink and exposing it to UV light for 5 min. The reference electrode was made by directly printing Ag/AgCl ink and allowing it to dry at room temperature. Subsequently, a UV insulating layer was applied to the SPEs surface to expose only the working area and electrode wire contacts. After these steps, 100 W plasma was used for 6 min to activate the SPEs. 1 mM HAuCl\u003csub\u003e4\u003c/sub\u003e\u0026middot;3H\u003csub\u003e2\u003c/sub\u003eO and 5 mM K\u003csub\u003e2\u003c/sub\u003ePtCl\u003csub\u003e4\u003c/sub\u003e solution in 0.5 M H\u003csub\u003e2\u003c/sub\u003eSO\u003csub\u003e4\u003c/sub\u003e were mixed 1:1 before use. The AuPt composite material was prepared by CV to improve the conductivity of the SPEs, and the scan rate was 0.02 V/s in -0.2 to 0.5 V for 15 cycles. Finally, the SPEs surface was rinsed with deionized water and dried at room temperature. The samples were stored at 4\u0026deg;C.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e4.7 Preparation of biosensors\u003c/h2\u003e \u003cp\u003eThe 2 \u0026micro;M thiolated capture probe H1/H1' was incubated with 1 mM TCEP for 30 min at room temperature in the dark to break with disulphide bonds. 5 \u0026micro;L of 2 \u0026micro;M of pre-sulfide-capture probe H1/H1' was dropped onto the surface of the pre-treated AuPt-SPEs and incubated overnight at 4\u0026deg;C. After washing with washing solution, 5 \u0026micro;L of 1 mM MCH was drip-added to the SPEs surface and incubated at 4\u0026deg;C for 1 h to obtain a neatly arranged DNA monolayer, and nonspecific binding sites were blocked. After washing and drying at room temperature, electrochemical biosensors were obtained. Subsequently, 5 \u0026micro;L of 2 mΜ Fc-labeled H2/H2' and 5 \u0026micro;L of hsa_circ _044235 and hsa_circ _000250 were dropped-added to the SPEs surface for CHA amplification.\u003c/p\u003e \u003cp\u003eDPV measurements were performed in 10 mM tris-HCl (pH 7.4) to obtain a DPV response signal. The measurement parameters were set as a scan rate of 0.02 V/s, a resting time of 20 s, and a scan voltage range of 0.2 to 0.6 V.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e4.8 Electrochemical characterization of AuPt-SPEs\u003c/h2\u003e \u003cp\u003eFor detailed analysis of the surface at each stage of the electrochemically modified SPEs, the CV and OCP-EIS methods were used in 0.5 M KCl solution containing 10 mM K\u003csub\u003e4\u003c/sub\u003eFe(CN)\u003csub\u003e6\u003c/sub\u003e/K\u003csub\u003e3\u003c/sub\u003eFe(CN)\u003csub\u003e6\u003c/sub\u003e (1:1). The voltage range was \u0026minus;\u0026thinsp;0.2 to 1.0 V, the scan rate was 0.02 V/s, 2 cycles, the sampling interval was 0.001 V, and the standing time was 20 s. OCP-EIS was performed with the following Settings: the initial potential was set to the measured OCP voltage with a frequency range of 0.1 to 100,000 Hz and an amplitude of 0.005 V. The Randles-Sevcik equation was used to calculate the active surface area of the SPEs. There was a linear relationship between the CV anode peak current (Ip) and the square root of the scan rate (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\text{v}}^{\\frac{\\text{1}}{\\text{2}}}\\)\u003c/span\u003e\u003c/span\u003e).\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:\\begin{array}{c}\\text{I}\\text{p}\\text{=}\\frac{\\text{2.69\u0026times;}{\\text{10}}^{\\text{5}}\\text{\u0026times;}{\\text{n}}^{\\frac{\\text{3}}{\\text{2}}}\\text{\u0026times;C\u0026times;A\u0026times;}{\\text{v}}^{\\frac{\\text{1}}{\\text{2}}}}{{\\text{D}}^{\\frac{\\text{1}}{\\text{2}}}}\\#（\\text{1}）\\end{array}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere Ip refers to the peak current (A), A is the surface active area (cm\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e), n is the number of transferred electrons (n\u0026thinsp;=\u0026thinsp;1), C is the concentration of [Fe(CN)]\u003csup\u003e3\u0026minus;/4\u0026minus;\u003c/sup\u003e (5\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e mol/cm\u003csup\u003e3\u003c/sup\u003e), D is the diffusion coefficient (6.7\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e cm\u003csup\u003e2\u003c/sup\u003e/s) of [Fe(CN)]\u003csup\u003e3\u0026minus;/4\u0026minus;\u003c/sup\u003e in 0.5 M KCI solution, and v is the potential sweep rate (V/s). The surface area of the SPEs was calculated from the slope of Ip vs. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\text{v}}^{\\frac{\\text{1}}{\\text{2}}}\\)\u003c/span\u003e\u003c/span\u003e. The slope of the bare SPEs and AuPt-SPEs was 20.702 and 128.03, respectively.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e4.9 Experimental details of Native -polyacrylamide gel electrophoresis\u003c/h2\u003e \u003cp\u003eFirstly, H1/H1' and H2/H2' were subjected to CHA reaction with hsa_circ_044235 and hsa_circ_000250 at 37\u0026deg;C for 2 h, respectively. In these reactions, a final concentration of 2 \u0026micro;M (H1/H1', H2/H2') of DNA was used, except for circRNAs, which was 100 nM. The product corresponding to each lane was then mixed with 6\u0026times;loading buffer. 0.5\u0026times;Tris-borate-EDTA buffer (TBE) was selected as the running buffer. The electrophoresis experiment was carried out in 15% natural polyacrylamide gel electrophoresis at a constant voltage of 120 V for 2 h, and finally stained with GelstainRed nucleic acid dye. Photographs were taken with a FireReader V10 Plus gel imaging system and finally processed with Photoshop software.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e4.10 ECMCDP fabrication and assembly\u003c/h2\u003e \u003cp\u003eThe power supply of the ECMCDP electronic system was regulated by a lithium-ion battery combined with a switched capacitor boost converter (PW510B-3.3V, PWCHIP). The system microcontroller unit (MCU) uses Bluetooth low energy chip NRF52832-QFAA-R (Microchip Technology), and the built-in analog-to-digital converter (ADC) to realize signal acquisition. Voltage programming was generated by the DAC module MCP4725AOT-E/CH (Microchip Technology) and applied to the sensor electrode by the potentiostat module LMV612 (Texas Instruments). The current signal was processed by the trans-impedance amplifier composed of MCP6031T-E/OT precision operational amplifier (Microchip Technology), collected by the ADC of MCU, and transmitted to the smartphone terminal through Bluetooth for real-time analysis. The current signal has its own trans-impedance amplifier channel according to the different detection substances.\u003c/p\u003e \u003cp\u003eA microfluidic chip mold for nucleic acid detection integrated with the chip was designed and prepared to form a PMMA mold. The PDMS prepolymer was blended with the curing agent in a 10:1 ratio. The mixture was then poured into the mold and a degassing process was carried out under vacuum for 15 min to eliminate air bubbles. Subsequently, the molds were cured at 85\u0026deg;C for 1 h. The PDMS microfluidic chip was cleaned by soaking in 75% ethanol and deionized water, respectively. The PDMS support layer and the electrode layer were connected with VHB glue and then connected to the FPCB, which was welded to the cell to complete the assembly.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e4.11 Detection of ECMCDP in clinical samples\u003c/h2\u003e \u003cp\u003eAfter sample collection, total RNA was extracted by RNA extraction kit, and RNA and H2/H2' were directly prestored on the microfluidic chip. Under the action of a peristaltic pump, the reaction was carried out on the surface of H1/H1'-SPEs, and ECMCDP collected the electrochemical signal after 37 ℃ for 2 h. The sensitivity and specificity of the circRNAs detection platform were calculated using the cut-off value of the maximum Youden index (Youden index\u0026thinsp;=\u0026thinsp;maximum value (sensitivity\u0026thinsp;+\u0026thinsp;specificity \u0026minus;\u0026thinsp;1)).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e4.12 Machine Learning\u003c/h2\u003e \u003cp\u003eAll analyses were performed using Python (version 3.5) with publicly available packages, including Pandas (v1.1.5), Scikit-learn (v0.24.2), Keras (v2.4.3), and Tensorflow (v2.4.1). The source code for data preprocessing, model training, and visualization is available under the MIT License at Zenodo (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5281/zenodo.152039487\u003c/span\u003e\u003cspan address=\"10.5281/zenodo.152039487\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and GitHub (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/Lamzsen/BCclassifier\u003c/span\u003e\u003cspan address=\"https://github.com/Lamzsen/BCclassifier\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, release v1.0.1).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e4.13 Statistical analysis\u003c/h2\u003e \u003cp\u003eStatistical analyses of all data were performed using Graph Pad Prism 9.0 software. Spearman correlation analysis was performed using IBM SPSS Statistics 26. All statistical tests are indicated in the figure legends. LOD was calculated using the standard slope method:\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:\\begin{array}{c}\\text{L}\\text{O}\\text{D}\\text{=}\\frac{\\text{3\u0026delta;}}{\\text{s}}\\#（\\text{2}）\\end{array}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\text{\u0026delta;}\\)\u003c/span\u003e\u003c/span\u003e is the standard deviation of the intercept, and s is the slope of the standard curve.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eSupporting Information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Supporting Information is available free of charge at http://. or from the author.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCorresponding Author:\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e*\u003c/sup\u003eYong Xia: [email protected]\u003c/p\u003e\n\u003cp\u003e*Lei Mou: [email protected]\u003c/p\u003e\n\u003cp\u003e*Honghai Hong: [email protected]\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eXinyu Zhang: conceptualization, data collation, formal analysis, investigation, validation, writing-first draft, writing-review, and editing. Zixin Lin: Conceptualization, data management, visualization, writing-first draft, writing-review and editing. Yongmei Chen: Data management, formal analysis, visualization, writing-review and editing. Huanyu Zhou: Methodology, writing review and editing. Yuetao Zhang: Formal analysis, software, writing-review and editing. ZiJie Li: data management, formal analysis, writing-review and editing. Runlin lv: Resources, writing-review and editing. Zhiqi Li: Research, writing, review, editing. Rubing Xiong: Research, writing, review, editing. Yong Xia: Conceptualization, writing-review and editing, funding acquisition, supervision. Mou Lei: Conceptualization, writing-review and editing, funding acquisition, supervision. Honghai Hong: Concept, methodology, writing-review and editing, funding acquisition, supervision.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding Information:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank Guangzhou Municipal Science and Technology Foundation (2025A03J3757); Department of Science and Technology of Guangdong Province (2023A1515110357); Guangzhou Medical University (02-408-2203-2058, PX-66242684) for financial support.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data supporting the findings of this study are available in the paper and its Supplementary Information. Source data are provided in this paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing financial interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBray, F. \u003cem\u003eet al.\u003c/em\u003e Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. Ca-Cancer. J. Clin. 74, 229\u0026ndash;263 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, S. \u003cem\u003eet al.\u003c/em\u003e V2C-Driven Nanodelivery Platform Potentiates Synergistic Breast Cancer Therapy. ACS. Mater. Lett. 5, 3017\u0026ndash;3031 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoibl, S. \u003cem\u003eet al.\u003c/em\u003e Breast cancer. The Lancet. 397, 1750\u0026ndash;1769 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, B. \u003cem\u003eet al.\u003c/em\u003e Targeting non-coding RNAs to overcome cancer therapy resistance. Signal. Transduct. Tar. 7, 121 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, J. \u003cem\u003eet al.\u003c/em\u003e Circular RNA circRHOBTB3 represses metastasis by regulating the HuR-mediated mRNA stability of PTBP1 in colorectal cancer. Theranostics. 11, 7507\u0026ndash;7526 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGong, X. \u003cem\u003eet al.\u003c/em\u003e Circular RNA circEsyt2 regulates vascular smooth muscle cell remodeling via splicing regulation. J. Clin. Invest. 131 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConn, V. M. \u003cem\u003eet al.\u003c/em\u003e Circular RNA in cancer. Nat. Rev. Cancer. 24, 597\u0026ndash;613 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, B. \u003cem\u003eet al.\u003c/em\u003e An integrated electrochemical POCT platform for ultrasensitive circRNA detection towards hepatocellular carcinoma diagnosis. Biosens. Bioelectron. 192, 113500 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSha, L. \u003cem\u003eet al.\u003c/em\u003e Collaborative CRISPR-Cas System-Enabled Detection of Circulating Circular RNA for Reliable Monitoring of Acute Myocardial Infarction. Small. 20, 2402895 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDigby, B. \u003cem\u003eet al.\u003c/em\u003e Computational approaches and challenges in the analysis of circRNA data. Bmc. Genomics. 25, 527 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQian, J. \u003cem\u003eet al.\u003c/em\u003e A microdroplet SERS-RCA biosensor with enhanced specificity and reproducibility for profiling dual miRNAs in idiopathic pulmonary fibrosis diagnosis and monitoring. Chem. Eng. J. 482, 149012 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMohammadniaei, M. \u003cem\u003eet al.\u003c/em\u003e A non-enzymatic, isothermal strand displacement and amplification assay for rapid detection of SARS-CoV-2 RNA. Nat. Commun. 12, 5089 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao, Y. \u003cem\u003eet al.\u003c/em\u003e Isothermal Amplification of Nucleic Acids. Chem. Rev. 115, 12491\u0026ndash;12545 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, C. \u003cem\u003eet al.\u003c/em\u003e Application of Nanomaterials in Isothermal Nucleic Acid Amplification. Small. 18, 2102711 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, J. \u003cem\u003eet al.\u003c/em\u003e Applications of Catalytic Hairpin Assembly Reaction in Biosensing. Small. 15, 1902989 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, Y. \u003cem\u003eet al.\u003c/em\u003e A fully integrated, non-invasive, and multimodal wearable device for sweat stimulation, collection and multiple physiological signals analysis. Chem. Eng. J. 505, 159209 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYe, C. \u003cem\u003eet al.\u003c/em\u003e Nucleic acid-based wearable and implantable electrochemical sensors. Chem. Soc. Rev. 53, 7960\u0026ndash;7982 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, Y. S. \u003cem\u003eet al.\u003c/em\u003e Multisensor-integrated organs-on-chips platform for automated and continual in situ monitoring of organoid behaviors. \u003cem\u003ePNAS\u003c/em\u003e. 114, E2293-E2302 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng, W. \u003cem\u003eet al.\u003c/em\u003e Diagnosis of paediatric tuberculosis by optically detecting two virulence factors on extracellular vesicles in blood samples. Nat. Biomed. Eng. 6, 979\u0026ndash;991 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKashefi-Kheyrabadi, L. \u003cem\u003eet al.\u003c/em\u003e Rapid, multiplexed, and nucleic acid amplification-free detection of SARS-CoV-2 RNA using an electrochemical biosensor. Biosens. Bioelectron. 195, 113649 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDastidar, M. G. \u003cem\u003eet al.\u003c/em\u003e Emerging electrochemical nanoscale architectures for point-of-care detection of miRNA biomarkers. Chem. Eng. J. 508, 160903 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCobzariu, M. A. \u003cem\u003eet al.\u003c/em\u003e Enzyme-assisted isothermal amplification of nucleic acids on the electrode surface. Curr. Opin. Electroche. 40, 101322 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, K. \u003cem\u003eet al.\u003c/em\u003e A novel signal amplification label based on AuPt alloy nanoparticles supported by high-active carbon for the electrochemical detection of circulating tumor DNA. Anal. Chim. Acta. 1169, 338628 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Y. \u003cem\u003eet al.\u003c/em\u003e Fast and in-situ electrodeposition of MXene/AuNPs composite for multiplexed and sensitive detection of infectious biomarkers using an electrochemical biosensor. Microchem. J. 207, 112064 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTu, J. \u003cem\u003eet al.\u003c/em\u003e A wireless patch for the monitoring of C-reactive protein in sweat. Nat. Biomed. Eng. 7, 1293\u0026ndash;1306 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong, S. \u003cem\u003eet al.\u003c/em\u003e Photo-Assisted Robust Anti-Interference Self-Powered Biosensing of MicroRNA Based on Pt\u0026ndash;S Bonds and the Inorganic\u0026ndash;Organic Hybridization Strategy. Anal. Chem. 94, 1654\u0026ndash;1660 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQuan, K. \u003cem\u003eet al.\u003c/em\u003e Pt-Decorated Gold Nanoflares for High-Fidelity Phototheranostics: Reducing Side-Effects and Enhancing Cytotoxicity toward Target Cells. Angew. Chem. Int. Edit. 63, e202402881 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCao, X. \u003cem\u003eet al.\u003c/em\u003e Bimetallic AuPt nanochains: Synthesis and their application in electrochemical immunosensor for the detection of carcinoembryonic antigen. Biosens. Bioelectron. 39, 226\u0026ndash;230 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, B. \u003cem\u003eet al.\u003c/em\u003e Programmable CRISPR-Cas9 microneedle patch for long-term capture and real-time monitoring of universal cell-free DNA. Nat. Commun. 13, 3999 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Y. \u003cem\u003eet al.\u003c/em\u003e Multiplexed discrimination of SARS-CoV-2 variants via plasmonic-enhanced fluorescence in a portable and automated device. Nat. Biomed. Eng. 7, 1636\u0026ndash;1648 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, S. \u003cem\u003eet al.\u003c/em\u003e Present status of microfluidic PCR chip in nucleic acid detection and future perspective. Trac-Trend. Anal. Chem. 157, 116737 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao, D. \u003cem\u003eet al.\u003c/em\u003e Microfluidic chip and isothermal amplification technologies for the detection of pathogenic nucleic acid. J. Biol. Eng. 16, 33 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSwanson, K. \u003cem\u003eet al.\u003c/em\u003e From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment. Cell. 186, 1772\u0026ndash;1791 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, Z. \u003cem\u003eet al.\u003c/em\u003e Machine Learning-Aided Identification of Fecal Extracellular Vesicle microRNA Signatures for Noninvasive Detection of Colorectal Cancer. ACS. Nano. 19, 10013\u0026ndash;10025 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSimmel, F. C. \u003cem\u003eet al.\u003c/em\u003e Principles and Applications of Nucleic Acid Strand Displacement Reactions. Chem. Rev. 119, 6326\u0026ndash;6369 (2019).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Breast cancer, CircRNAs, Electrochemical biosensor, Point-of-Care detection, Machine learning ","lastPublishedDoi":"10.21203/rs.3.rs-6504183/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6504183/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe discovery and favourable detection of breast cancer biomarkers are significant for cancer diagnosis. Here, we show that hsa_circ_044235 and hsa_circ_000250 are effective biomarkers for breast cancer diagnosis. Moreover, we present an integrated electrochemical microfluidic circRNAs detection platform (ECMCDP) that combined gold platinum nanoparticles (AuPts)-modified screen-printed electrodes (SPEs), catalytic hairpin assembly (CHA), electrochemical microfluidic chip and a customed low-power electronic system for simultaneous detection of two breast cancer-associated circRNAs. The limit of detection (LOD) were 0.12 fM and 0.1 fM, and the diagnostic accuracy were 92.50% and 88.75% in clinical blood samples, respectively. The platform was validated using paired pre-/post-operative blood and tissue samples. Combined with five machine learning-based diagnostic models, the ensemble diagnosis model achieved a high accuracy of 93.75%. This work aims to identify novel breast cancer biomarkers and establish an innovative circRNAs detection platform to improve breast cancer diagnosis and support clinical prognosis assessment.\u003c/p\u003e","manuscriptTitle":"A Novel Machine Learning-enhanced Microfluidic CircRNAs Detection Platform for Breast Cancer Precision Diagnosis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-07 07:09:23","doi":"10.21203/rs.3.rs-6504183/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"e73c7796-19a0-404c-b2a2-2b66056852d3","owner":[],"postedDate":"May 7th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":47560786,"name":"Physical sciences/Nanoscience and technology/Nanobiotechnology/Biosensors"},{"id":47560787,"name":"Biological sciences/Cancer/Breast cancer"},{"id":47560788,"name":"Physical sciences/Nanoscience and technology/Nanobiotechnology/Microfluidics"}],"tags":[],"updatedAt":"2025-05-15T06:45:30+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-07 07:09:23","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6504183","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6504183","identity":"rs-6504183","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}