Explainable Deep Learning–Based Prediction of hERG Channel Blockage: A Cardio-Oncology Perspective

Abstract

Cardiotoxicity induced by human Ether-a-go-go-Related Gene (hERG) channel blockade remains a critical safety liability, particularly in cardio-oncology, where antineoplastic agents pose severe proarrhythmic risks. This study developed an explainable deep learning-based QSAR framework for the multidimensional prediction of hERG-mediated cardiotoxicity. A primary dataset of 22,213 compounds was used to train a deep neural network (DNN), utilizing RDKit descriptors, MACCS keys, and Morgan fingerprints individually and in hybrid combinations. The integrated DNN model achieved an exceptional ROC-AUC of 0.9801 during the training phase. For external validation, an independent oncology test set of 348 antineoplastic agents was evaluated. The results revealed that the high structural complexity of chemotherapeutics limits generalizability, highlighting the multifactorial nature of oncology-related cardiotoxicity. Crucially, explainable artificial intelligence (XAI) analyses demonstrated that the model's decisions are firmly grounded in pharmacologically coherent principles, identifying lipophilicity (MolLogP), polar surface area, and specific topological motifs as primary toxicity determinants. Overall, this explainable QSAR approach offers a transparent, mechanistically interpretable decision-support tool for early-stage cardiotoxicity screening, bridging the gap between computational predictions and clinical cardio-oncology safety assessments.

Keywords

• hERG
• Cardiotoxicity
• QSAR
• Deep learning
• Explainable artificial intelligence
• Anticancer drugs

Supporting Institution

Sakarya University of Applied Sciences

Ethical Statement

Ethics committee approval was not required for this study because it involved no animal or human subjects.

Thanks

This study was derived from the master’s thesis of Tuba Seyhan completed at Sakarya University of Applied Sciences.

References

1. Aires-de-Sousa, J. (2023). GUIDEMOL: A Python graphical user interface for molecular descriptors based on RDKit. Molecular Informatics, 42(8-9), Article e202300190. https://doi.org/10.1002/minf.202300190
2. Almahmood, M., Najadat, H., Alzubi, D., Abualigah, L., Zitar, R. A., Abualigah, S., & Al-Tawil, M. (2023). Predictive model of psychoactive drugs consumption using classification machine learning algorithms. Applied and Computational Engineering, 8(1), 853–858. https://doi.org/10.54254/2755-2721/8/20230097
3. Arab, I., Laukens, K., & Bittremieux, W. (2024). Semisupervised learning to boost hERG, Nav1.5, and Cav1.2 cardiac ion channel toxicity prediction by mining a large unlabeled small molecule data set. Journal of Chemical Information and Modeling, 64(16), 6410–6420. https://doi.org/10.1021/acs.jcim.4c01102
4. Ayres, L. B., Weavil, G., Alhoubani, M., Guinati, B. G. S., & Garcia, C. D. (2023). Big data for a deep problem: Understanding the formation of NADES through comprehensive chemical analysis and RDKit. Journal of Molecular Liquids, 389, Article 122891. https://doi.org/10.1016/j.molliq.2023.122891
5. Banerjee, A., & Roy, K. (2023). Machine-learning-based similarity meets traditional QSAR: q-RASAR for the enhancement of the external predictivity and detection of prediction confidence outliers in an hERG toxicity dataset. Chemometrics and Intelligent Laboratory Systems, 237, Article 104829. https://doi.org/10.1016/j.chemolab.2023.104829
6. Bento, A. P., Hersey, A., Félix, E., Landrum, G., Gaulton, A., Atkinson, F., Bellis, L. J., de Veij, M., & Overington, J. P. (2020). An open source chemical structure curation pipeline using RDKit. Journal of Cheminformatics, 12, Article 51. https://doi.org/10.1186/s13321-020-00456-1
7. Cai, C., Guo, P., Zhou, Y., Zhou, J., Wang, Q., Zhang, F., Jiao, J., & Ding, X. (2019). Deep learning-based prediction of drug-induced cardiotoxicity. Journal of Chemical Information and Modeling, 59(3), 1073–1084. https://doi.org/10.1021/acs.jcim.8b00769
8. ChEMBL database. (b.t.). European Bioinformatics Institute. https://www.ebi.ac.uk/chembl/

9. Cucos, A. M., & Iantovics, L. B. (2024). Comparative study of random forest, gradient boosted trees, feedforward neural networks and convolutional neural networks using fingerprints and molecular descriptors for adverse drug reaction prediction. Procedia Computer Science, 246, 1895–1904. https://doi.org/10.1016/j.procs.2024.09.702
10. Curigliano, G., Cardinale, D., Suter, T., Plataniotis, G., De Azambuja, E., Sandri, M. T., Criscitiello, C., Goldhirsch, A., & ESMO Guidelines Working Group. (2012). Cardiovascular toxicity induced by chemotherapy, targeted agents and radiotherapy: ESMO clinical practice guidelines. Annals of Oncology, 23(Suppl. 7), vii155–vii166. https://doi.org/10.1093/annonc/mds293
11. Daniels, R. W., Kramer, J. W., Abernathy, M. M., Best, D. D., Brimecombe, J. C., & Leishman, D. J. (2025). Legacy hERG study data in contemporary integrated risk assessments: Mitigating the risk of protocol and positive control article differences. Journal of Pharmacological and Toxicological Methods, 135, Article 108387. https://doi.org/10.1016/j.vascn.2025.108387
12. Deng, J., Yang, Z., Wang, H., Ojima, I., Samaras, D., & Wang, F. (2023). A systematic study of key elements underlying molecular property prediction. Nature Communications, 14, Article 6395. https://doi.org/10.1038/s41467-023-41948-6
13. Du , F., Babcock, J. J., Yu, H., Zou, B., & Li, M. (2015). Global analysis reveals families of chemical motifs enriched for hERG inhibitors. PLOS ONE, 10(2), Article e0118324. https://doi.org/10.1371/journal.pone.0118324
14. European Medicines Agency. (2005). ICH E14: The clinical evaluation of QT/QTc interval prolongation and proarrhythmic potential for non-anti-arrhythmic drugs (CPMP/ICH/2/04). European Medicines Agency.
15. Friesacher, H. R., Svensson, E., Winiwarter, S., Mervin, L., Arany, A., & Engkvist, O. (2025). Temporal distribution shift in real-world pharmaceutical data: Implications for uncertainty quantification in QSAR models. Artificial Intelligence in the Life Sciences, 8, Article 100132. https://doi.org/10.1016/j.ailsci.2025.100132
16. Garrido, A., Lepailleur, A., Mignani, S. M., Dallemagne, P., & Rochais, C. (2020). hERG toxicity assessment: Useful guidelines for drug design. European Journal of Medicinal Chemistry, 195, Article 112290. https://doi.org/10.1016/j.ejmech.2020.112290
17. Hasan, M. M., Tarkhaneh, O., Bungay, S. D., Poirier, R. A., & Islam, S. M. (2025). Predicting HOMO–LUMO gaps using Hartree–Fock calculated data and machine learning models. Journal of Chemical Information and Modeling, 65(3), 9497–9515. https://doi.org/10.1021/acs.jcim.5c01412
18. He , S., Moutaoufik, M. T., Islam, S., Persad, A., Wu, A., Aly, K. A., Figeys, D., & Babu, M. (2020). hERG channel and cancer: A mechanistic review of carcinogenic processes and therapeutic potential. Biochimica et Biophysica Acta (BBA) – Reviews on Cancer, 1873(2), Article 188355. https://doi.org/10.1016/j.bbcan.2020.188355
19. Iftkhar, S., De Sá, A. G. C., Velloso, J. P. L., Aljarf, R., Pires, D. E. V., & Ascher, D. B. (2022). CardioToxCSM: A web server for predicting cardiotoxicity of small molecules. Journal of Chemical Information and Modeling, 62(20), 4827–4836. https://doi.org/10.1021/acs.jcim.2c00822
20. Kalyaanamoorthy, S., & Barakat, K. H. (2018). Development of safe drugs: The hERG challenge. Medical Research Reviews, 38(2), 525–555. https://doi.org/10.1002/med.21445
21. Karim, A., Lee, M., Balle, T., & Sattar, A. (2021). CardioTox net: A robust predictor for hERG channel blockade based on deep learning meta-feature ensembles. Journal of Chemical Information and Modeling, 13, Article 41. https://doi.org/10.1186/s13321-021-00541-z
22. Kim, D., Jeong, J., & Choi, J. (2024). Identification of optimal machine learning algorithms and molecular fingerprints for explainable toxicity prediction models using ToxCast/Tox21 bioassay data. ACS Omega, 9(36), 37934–37941. https://doi.org/10.1021/acsomega.4c04474
23. Kim, S.-Y., Lee, K.-H., Park, S. Y., Han, S. J., Ahn, K.-I., & Hwang, S. W. (2022). Interfacing between MAAP and MACCS to perform radiological consequence analysis. Nuclear Engineering and Technology, 54(5), 1516–1525. https://doi.org/10.1016/j.net.2021.10.001
24. Kojic, A., Moslehi, J., Ky, B., & Wu, J. C. (2025). Cardiometabolic disease and cardio-oncology: Insights from iPSC models and tissue engineering. Cell Reports Medicine, Article 102261. https://doi.org/10.1016/j.xcrm.2025.102261
25. Koutsoukis, A., Ntalianis, A., Repasos, E., Kastritis, E., Dimopoulos, M. A., & Paraskevaidis, I. (2018). Cardio-oncology: A focus on cardiotoxicity. European Cardiology Review, 13(1), 64–69. https://doi.org/10.15420/ecr.2017:17:2
26. Kratz, J. M., Schuster, D., Edtbauer, M., Saxena, P., Mair, C. E., Kirchebner, J., Langer, T., & Rollinger, J. M. (2014). Experimentally validated hERG pharmacophore models as cardiotoxicity prediction tools. Journal of Chemical Information and Modeling, 54(10), 2887–2901. https://doi.org/10.1021/ci5001955
27. Lee, H. M., Yu, M. S., Kazmi, S. R., Oh, S. Y., Rhee, K. H., Bae, M. A., Shin, M. S., & Chang, S. Y. (2019). Computational determination of hERG-related cardiotoxicity of drug candidates. BMC Bioinformatics, 20, Article 250. https://doi.org/10.1186/s12859-019-2814-5
28. Lovrić, M., Molero, J. M., & Kern, R. (2019). PySpark and RDKit: Moving towards big data in cheminformatics. Molecular Informatics, 38(6), Article e1800082. https://doi.org/10.1002/minf.201800082
29. Mastrolorito, F., Togo, M. V., Gambacorta, N., Trisciuzzi, D., Giannuzzi, V., Bonifazi, F., Altomare, C. D., Mangiatordi, G. F., & Nicolotti, O. (2024). TISBE: A public web platform for the consensus-based explainable prediction of developmental toxicity. Chemical Research in Toxicology, 37(2), 323–339. https://doi.org/10.1021/acs.chemrestox.3c00310
30. Mazanetz, M. P., Marmon, R. J., Reisser, C. B. T., & Morao, I. (2012). Drug discovery applications for KNIME: An open source data mining platform. Current Topics in Medicinal Chemistry, 12(18), 1965–1979. https://doi.org/10.2174/156802612804910331
31. Morelli, M. B., Bongiovanni, C., Da Pra, S., Miano, C., Sacchi, F., Lauriola, M., D'Uva, G., & Alviano, F. (2022). Cardiotoxicity of anticancer drugs: Molecular mechanisms and strategies for cardioprotection. Frontiers in Cardiovascular Medicine, 9, Article 847012. https://doi.org/10.3389/fcvm.2022.847012
32. Morgan, H. L. (2002). The generation of a unique machine description for chemical structures: A technique developed at Chemical Abstracts Service. Journal of Chemical Documentation, 5(2), 107–113. https://doi.org/10.1021/C160017A018
33. Orosz, Á., Héberger, K., & Rácz, A. (2022). Comparison of descriptor- and fingerprint sets in machine learning models for ADME–Tox targets. Frontiers in Chemistry, 10, Article 852893. https://doi.org/10.3389/fchem.2022.852893
34. Pala, M. A. (2026). XP-GCN: Extreme learning machines and parallel graph convolutional networks for high-throughput prediction of blood–brain barrier penetration based on feature fusion. Computational Biology and Chemistry, 120, Article 108755. https://doi.org/10.1016/j.compbiolchem.2025.108755
35. Parker, J. A., Fung, E. S., Trejo-Martin, A., Liang, L., Gibbs, K., Bandara, S., & Borghoff, S. J. (2022). The utility of hERG channel inhibition data in the derivation of occupational exposure limits. Regulatory Toxicology and Pharmacology, 134, Article 105224. https://doi.org/10.1016/j.yrtph.2022.105224
36. Patanè, S. (2014). hERG-targeted therapy in both cancer and cardiovascular system with cardiovascular drugs. International Journal of Cardiology, 176(3), 1082–1085. https://doi.org/10.1016/j.ijcard.2014.07.129
37. Pillozzi, S., Brizzi, M. F., Bernabei, P. A., Bartolozzi, B., Caporale, R., Basile, V., Boddi, V., Pegoraro, L., Becchetti, A., & Arcangeli, A. (2007). VEGFR-1 (FLT-1), β1 integrin, and hERG K⁺ channel form a macromolecular signaling complex in acute myeloid leukemia: Role in cell migration and clinical outcome. Blood, 110(4), 1238–1250. https://doi.org/10.1182/blood-2006-02-003772
38. Ref-MACCS keys implementation (Chem/MACCSkeys.py). (b.t.). GitHub. https://github.com/rdkit/rdkit/blob/master/rdkit/Chem/MACCSkeys.py
39. Rogers, D., & Hahn, M. (2010). Extended-connectivity fingerprints. Journal of Chemical Information and Modeling, 50(5), 742–754. https://doi.org/10.1021/ci100050t
40. Seal, S., Mahale, M., García-Ortegón, M., Joshi, C. K., Hosseini-Gerami, L., Beatson, A., & Bender, A. (2025). Machine learning for toxicity prediction using chemical structures: Pillars for success in the real world. Chemical Research in Toxicology, 38(5), 759–807. https://doi.org/10.1021/acs.chemrestox.5c00033
41. Sharma, B., Chenthamarakshan, V., Dhurandhar, A., Pereira, S., Hendler, J. A., Dordick, J. S., & Payani, A. (2023). Accurate clinical toxicity prediction using multi-task deep neural nets and contrastive molecular explanations. Scientific Reports, 13, Article 4908. https://doi.org/10.1038/s41598-023-31169-8
42. Shayimu, P., Yusufu, A., Rehemutula, A., Redati, D., Jiapaer, R., & Tuerdi, R. (2021). MTBP promoted the proliferation, migration and invasion of colon cancer cells by activating the expression of ZEB2. Animal Cells and Systems, 25(3), 152–160. https://doi.org/10.1080/19768354.2021.1938218
43. Sliwoski, G., Kothiwale, S., Meiler, J., & Lowe, E. W. (2014). Computational methods in drug discovery. Pharmacological Reviews, 66(1), 336–386. https://doi.org/10.1124/pr.112.007336
44. Tetterton-Kellner, J., Jensen, B. C., & Nguyen, J. (2024). Navigating cancer therapy–induced cardiotoxicity: From pathophysiology to treatment innovations. Advanced Drug Delivery Reviews, 211, Article 115361. https://doi.org/10.1016/j.addr.2024.115361
45. Thai, K. M., & Ecker, G. F. (2008). A binary QSAR model for classification of hERG potassium channel blockers. Bioorganic & Medicinal Chemistry, 16(7), 4107–4119. https://doi.org/10.1016/j.bmc.2008.01.017
46. Tsantili-Kakoulidou, A., & Demopoulos, V. J. (2021). Drug-like properties and fraction lipophilicity index as a combined metric. ADMET & DMPK, 9(3), Article 177. https://doi.org/10.5599/admet.1022
47. Uddin, I., Awan, H. H., Khalid, M., Khan, S., Akbar, S., Sarker, M. R., & Shafi, M. (2024). A hybrid residue-based sequential encoding mechanism with XGBoost improved ensemble model for identifying 5-hydroxymethylcytosine modifications. Scientific Reports, 14, Article 71568. https://doi.org/10.1038/s41598-024-71568-z
48. Walters, W. P. (2025). Applications of machine learning in quantitative structure–activity relationships (QSAR). Reference Module in Chemistry, Molecular Sciences and Chemical Engineering, 1–15. Elsevier. https://doi.org/10.1016/B978-0-443-29808-0.00032-7
49. Wan, J., Xu, H., Ju, J., Chen, Y., Zhang, H., Qi, L., & Wang, L. (2024). Inhibition of hERG by ESEE suppresses the progression of colorectal cancer. Translational Oncology, 50, Article 102137. https://doi.org/10.1016/j.tranon.2024.102137
50. Wei, X., Sumithran, S. P., Deaciuc, A. G., Burton, H. R., Bush, L. P., Dwoskin, L. P., & Crooks, P. A. (2005). Identification and synthesis of novel alkaloids from the root system of Nicotiana tabacum: Affinity for neuronal nicotinic acetylcholine receptors. Life Sciences, 78(5), 495–505. https://doi.org/10.1016/j.lfs.2005.09.009
51. Wen, M., Zhang, Z., Niu, S., Sha, H., Yang, R., Yun, Y., & Lu, H. (2017). Deep-learning-based drug–target interaction prediction. Journal of Proteome Research, 16(4), 1401–1409. https://doi.org/10.1021/acs.jproteome.6b00618
52. Wu, L., Huang, R., Tetko, I. V., Xia, Z., Xu, J., & Tong, W. (2021). Trade-off predictivity and explainability for machine-learning powered predictive toxicology: An in-depth investigation with Tox21 data sets. Chemical Research in Toxicology, 34(2), 541–549. https://doi.org/10.1021/acs.chemrestox.0c00373
53. Xu, C., Xu, Y., Hu, Z., Zhao, X., Xie, W., Chen, W., & Zhang, Y. (2025). Unveiling optimal molecular features for hERG insights with automatic machine learning. Journal of Pharmaceutical Analysis, Article 101411. https://doi.org/10.1016/j.jpha.2025.101411
54. Yang, H., Xiu, J., Yan, W., Liu, K., Cui, H., Wang, Z., & Han, W. (2025). Large language models as tools for molecular toxicity prediction: AI insights into cardiotoxicity. Journal of Chemical Information and Modeling, 65(6), 2268–2282. https://doi.org/10.1021/acs.jcim.4c01371
55. Ylipää, E., Chavan, S., Bånkestad, M., Broberg, J., Glinghammar, B., Norinder, U., & Spjuth, O. (2023). hERG-toxicity prediction using traditional machine learning and advanced deep learning techniques. Current Research in Toxicology, 5, Article 100121. https://doi.org/10.1016/j.crtox.2023.100121
56. Zhong, S., & Guan, X. (2023). Count-based Morgan fingerprint: A more efficient and interpretable molecular representation in developing machine learning-based predictive regression models for water contaminants’ activities and properties. Environmental Science & Technology, 57(45), 18193–18202. https://doi.org/10.1021/acs.est.3c02198