Activity Cliffs and Machine Learning Generalization in QSAR: Why Similar Molecules Remain Difficult to Predict

Authors

  • Teuku Rizky Noviandy Universitas Abulyatama
  • Mohsina Patwekar Department of Pharmacology, College of Medicine, Gulf Medical University, Ajman, United Arab Emirates
  • Faheem Patwekar Department of Pharmacognosy, Luqman College of Pharmacy, Gulbarga, India

DOI:

https://doi.org/10.60084/hjas.v4i2.462

Keywords:

Activity cliffs, QSAR, Molecular representation, Drug discovery

Abstract

Machine learning has substantially expanded the predictive capability of quantitative structure-activity relationship (QSAR) modeling, yet reliable generalization remains difficult in chemically challenging regions. Activity cliffs-pairs of highly similar molecules that exhibit large differences in biological activity-represent a particularly stringent test because they violate the local smoothness assumed by many similarity-based and representation-learning approaches. This narrative review examines why activity cliffs remain difficult for contemporary QSAR models and how they expose limitations that may be concealed by conventional global performance metrics. We discuss the definition, detection, and chemical origins of activity cliffs; their relationship to discontinuities in structure-activity landscapes; and the influence of molecular representation, training-test partitioning, structural redundancy, and data leakage on apparent model performance. Evidence from fingerprint-based machine learning, deep neural networks, graph neural networks, and recent molecular foundation models indicates that increasing model complexity does not consistently resolve activity-cliff prediction. Models with strong overall QSAR performance may still exhibit substantially larger errors or poor sensitivity for cliff-forming compounds. Recent approaches that explicitly incorporate activity-cliff information through pairwise, contrastive, or activity-aware representation learning provide a promising alternative to conventional compound-level regression. We therefore propose an activity-cliff-aware QSAR perspective integrating rigorous bioactivity curation, activity-landscape characterization, leakage-resistant validation, global and cliff-specific evaluation, applicability-domain analysis, uncertainty quantification, and cliff-aware learning objectives. Activity cliffs should ultimately be regarded not merely as problematic outliers, but as information-rich stress tests of molecular machine-learning generalization and local structure-activity understanding.

Downloads

Download data is not yet available.

References

  1. Hansch, C., and Fujita, T. (1964). P -σ-π Analysis. A Method for the Correlation of Biological Activity and Chemical Structure, Journal of the American Chemical Society, Vol. 86, No. 8, 1616–1626. doi:10.1021/ja01062a035. DOI: https://doi.org/10.1021/ja01062a035
  2. Breiman, L. (2001). Random Forests, Machine Learning, Vol. 45, No. 1, 5–32. doi:10.1023/A:1010933404324. DOI: https://doi.org/10.1023/A:1010933404324
  3. Cortes, C., and Vapnik, V. (1995). Support-Vector Networks, Machine Learning, Vol. 20, No. 3, 273–297. doi:10.1007/BF00994018. DOI: https://doi.org/10.1023/A:1022627411411
  4. Wu, Z., Ramsundar, B., Feinberg, E. N., Gomes, J., Geniesse, C., Pappu, A. S., Leswing, K., and Pande, V. (2018). MoleculeNet: A Benchmark for Molecular Machine Learning, Chemical Science, Vol. 9, No. 2, 513–530. doi:10.1039/C7SC02664A. DOI: https://doi.org/10.1039/C7SC02664A
  5. Mayr, A., Klambauer, G., Unterthiner, T., Steijaert, M., Wegner, J. K., Ceulemans, H., Clevert, D.-A., and Hochreiter, S. (2018). Large-Scale Comparison of Machine Learning Methods for DrugTarget Prediction on ChEMBL, Chemical Science, Vol. 9, No. 24, 5441–5451. doi:10.1039/C8SC00148K. DOI: https://doi.org/10.1039/C8SC00148K
  6. Sheridan, R. P. (2013). Time-Split Cross-Validation as a Method for Estimating the Goodness of Prospective Prediction., Journal of Chemical Information and Modeling, Vol. 53, No. 4, 783–790. doi:10.1021/ci400084k. DOI: https://doi.org/10.1021/ci400084k
  7. Wallach, I., and Heifets, A. (2018). Most Ligand-Based Classification Benchmarks Reward Memorization Rather than Generalization, Journal of Chemical Information and Modeling, Vol. 58, No. 5, 916–932. doi:10.1021/acs.jcim.7b00403. DOI: https://doi.org/10.1021/acs.jcim.7b00403
  8. Maggiora, G. M. (2006). On Outliers and Activity Cliffs—Why QSAR Often Disappoints, Journal of Chemical Information and Modeling, Vol. 46, No. 4, 1535–1535. doi:10.1021/ci060117s. DOI: https://doi.org/10.1021/ci060117s
  9. Stumpfe, D., Hu, Y., Dimova, D., and Bajorath, J. (2014). Recent Progress in Understanding Activity Cliffs and Their Utility in Medicinal Chemistry, Journal of Medicinal Chemistry, Vol. 57, No. 1, 18–28. doi:10.1021/jm401120g. DOI: https://doi.org/10.1021/jm401120g
  10. Stumpfe, D., Hu, H., and Bajorath, J. (2019). Evolving Concept of Activity Cliffs, ACS Omega, Vol. 4, No. 11, 14360–14368. doi:10.1021/acsomega.9b02221. DOI: https://doi.org/10.1021/acsomega.9b02221
  11. Sheridan, R. P., Karnachi, P., Tudor, M., Xu, Y., Liaw, A., Shah, F., Cheng, A. C., Joshi, E., Glick, M., and Alvarez, J. (2020). Experimental Error, Kurtosis, Activity Cliffs, and Methodology: What Limits the Predictivity of Quantitative Structure–Activity Relationship Models?, Journal of Chemical Information and Modeling, Vol. 60, No. 4, 1969–1982. doi:10.1021/acs.jcim.9b01067. DOI: https://doi.org/10.1021/acs.jcim.9b01067
  12. Golbraikh, A., Muratov, E., Fourches, D., and Tropsha, A. (2014). Data Set Modelability by QSAR, Journal of Chemical Information and Modeling, Vol. 54, No. 1, 1–4. doi:10.1021/ci400572x. DOI: https://doi.org/10.1021/ci400572x
  13. van Tilborg, D., Alenicheva, A., and Grisoni, F. (2022). Exposing the Limitations of Molecular Machine Learning with Activity Cliffs, Journal of Chemical Information and Modeling, Vol. 62, No. 23, 5938–5951. doi:10.1021/acs.jcim.2c01073. DOI: https://doi.org/10.1021/acs.jcim.2c01073
  14. Tamura, S., Miyao, T., and Bajorath, J. (2023). Large-Scale Prediction of Activity Cliffs Using Machine and Deep Learning Methods of Increasing Complexity, Journal of Cheminformatics, Vol. 15, No. 1, 4. doi:10.1186/s13321-022-00676-7. DOI: https://doi.org/10.1186/s13321-022-00676-7
  15. Dablander, M., Hanser, T., Lambiotte, R., and Morris, G. M. (2023). Exploring QSAR Models for Activity-Cliff Prediction, Journal of Cheminformatics, Vol. 15, No. 1, 47. doi:10.1186/s13321-023-00708-w. DOI: https://doi.org/10.1186/s13321-023-00708-w
  16. Development, O. for E. C. and. (2014). Guidance Document on the Validation of (Quantitative) Structure-Activity Relationship [(Q)SAR] Models, OECD Series on Testing and Assessment No. 69, OECD, Paris. doi:10.1787/9789264085442-en. DOI: https://doi.org/10.1787/9789264085442-en
  17. Cui, C., Su, X., Zhang, Z., Velez-Arce, A., Wang, J., Shi, X., Zhang, Y., Wu, J., Chen, Y. Z., Zitnik, M., and Shen, W. (2026). Activity-Cliff Awareness Enables Robust Graph Learning for Molecular Property Prediction, Nature Communications, Vol. 17, No. 1, 9376. doi:10.1038/s41467-026-75713-2. DOI: https://doi.org/10.1038/s41467-026-75713-2
  18. Guha, R., and Van Drie, J. H. (2008). Structure−Activity Landscape Index: Identifying and Quantifying Activity Cliffs, Journal of Chemical Information and Modeling, Vol. 48, No. 3, 646–658. doi:10.1021/ci7004093. DOI: https://doi.org/10.1021/ci7004093
  19. Wassermann, A. M., Wawer, M., and Bajorath, J. (2010). Activity Landscape Representations for Structure−Activity Relationship Analysis, Journal of Medicinal Chemistry, Vol. 53, No. 23, 8209–8223. doi:10.1021/jm100933w. DOI: https://doi.org/10.1021/jm100933w
  20. Rogers, D., and Hahn, M. (2010). Extended-Connectivity Fingerprints, Journal of Chemical Information and Modeling, Vol. 50, No. 5, 742–754. doi:10.1021/ci100050t. DOI: https://doi.org/10.1021/ci100050t
  21. Hu, X., Hu, Y., Vogt, M., Stumpfe, D., and Bajorath, J. (2012). MMP-Cliffs: Systematic Identification of Activity Cliffs on the Basis of Matched Molecular Pairs, Journal of Chemical Information and Modeling, Vol. 52, No. 5, 1138–1145. doi:10.1021/ci3001138. DOI: https://doi.org/10.1021/ci3001138
  22. Hussain, J., and Rea, C. (2010). Computationally Efficient Algorithm to Identify Matched Molecular Pairs (MMPs) in Large Data Sets, Journal of Chemical Information and Modeling, Vol. 50, No. 3, 339–348. doi:10.1021/ci900450m. DOI: https://doi.org/10.1021/ci900450m
  23. Hu, Y., and Bajorath, J. (2012). Exploration of 3D Activity Cliffs on the Basis of Compound Binding Modes and Comparison of 2D and 3D Cliffs, Journal of Chemical Information and Modeling, Vol. 52, No. 3, 670–677. doi:10.1021/ci300033e. DOI: https://doi.org/10.1021/ci300033e
  24. Hu, H., and Bajorath, J. (2020). Activity Cliffs Produced by Single-Atom Modification of Active Compounds: Systematic Identification and Rationalization Based on X-Ray Structures, European Journal of Medicinal Chemistry, Vol. 207, 112846. doi:10.1016/j.ejmech.2020.112846. DOI: https://doi.org/10.1016/j.ejmech.2020.112846
  25. Stumpfe, D., and Bajorath, J. (2012). Frequency of Occurrence and Potency Range Distribution of Activity Cliffs in Bioactive Compounds, Journal of Chemical Information and Modeling, Vol. 52, No. 9, 2348–2353. doi:10.1021/ci300288f. DOI: https://doi.org/10.1021/ci300288f
  26. Bemis, G. W., and Murcko, M. A. (1996). The Properties of Known Drugs. 1. Molecular Frameworks, Journal of Medicinal Chemistry, Vol. 39, No. 15, 2887–2893. doi:10.1021/jm9602928. DOI: https://doi.org/10.1021/jm9602928
  27. Yang, K., Swanson, K., Jin, W., Coley, C., Eiden, P., Gao, H., Guzman-Perez, A., Hopper, T., Kelley, B., Mathea, M., Palmer, A., Settels, V., Jaakkola, T., Jensen, K., and Barzilay, R. (2019). Analyzing Learned Molecular Representations for Property Prediction, Journal of Chemical Information and Modeling, Vol. 59, No. 8, 3370–3388. doi:10.1021/acs.jcim.9b00237. DOI: https://doi.org/10.1021/acs.jcim.9b00237
  28. Alshehade, S. A., Al Jabal, G., and Jebril, I. H. (2026). Systematic Multi-Target QSAR Benchmarking: Machine Learning Algorithms, Molecular Descriptors, and Validation, Journal of Chemical Information and Modeling, Vol. 66, No. 15, 8923–8941. doi:10.1021/acs.jcim.6c01237. DOI: https://doi.org/10.1021/acs.jcim.6c01237
  29. Kötter, A., Singh, K., Matter, H., Hessler, G., and Grebner, C. (2025). Upgrading Reliability in Molecular Property Prediction by Robust Quantification of Uncertainty from Machine Learning Models, Journal of Chemical Information and Modeling, Vol. 65, No. 20, 10819–10831. doi:10.1021/acs.jcim.5c00464. DOI: https://doi.org/10.1021/acs.jcim.5c00464
  30. López‐Pérez, K., and Miranda‐Quintana, R. A. (2025). Extended Activity Cliffs‐Driven Approaches on Data Splitting for the Study of Bioactivity Machine Learning Predictions, Molecular Informatics, Vol. 44, No. 1, e202400054. doi:10.1002/minf.202400054. DOI: https://doi.org/10.1002/minf.202400054
  31. Hu, W., Liu, B., Gomes, J., Zitnik, M., Liang, P., Pande, V., and Leskovec, J. (2019). Strategies for Pre-Training Graph Neural Networks, ArXiv Preprint ArXiv:1905.12265.
  32. Honda, S., Shi, S., and Ueda, H. R. (2019). Smiles Transformer: Pre-Trained Molecular Fingerprint for Low Data Drug Discovery, ArXiv Preprint ArXiv:1911.04738.
  33. Aldeghi, M., Graff, D. E., Frey, N., Morrone, J. A., Pyzer-Knapp, E. O., Jordan, K. E., and Coley, C. W. (2022). Roughness of Molecular Property Landscapes and Its Impact on Modellability, Journal of Chemical Information and Modeling, Vol. 62, No. 19, 4660–4671. doi:10.1021/acs.jcim.2c00903. DOI: https://doi.org/10.1021/acs.jcim.2c00903
  34. Zhao, D., Zhu, Y., Wu, Z., Wan, Y., Liu, X., Li, S., Xu, H., Hou, T., and Hsieh, C.-Y. (2026). Revisiting ADMET Prediction Reliability under Real-World Challenges in the Foundation Model Era, Journal of Cheminformatics, Vol. 18, No. 1, 95. doi:10.1186/s13321-026-01217-2. DOI: https://doi.org/10.1186/s13321-026-01217-2

Downloads

Published

2026-09-30

How to Cite

Noviandy, T. R., Patwekar, M. and Patwekar, F. (2026) “Activity Cliffs and Machine Learning Generalization in QSAR: Why Similar Molecules Remain Difficult to Predict”, Heca Journal of Applied Sciences, 4(2), pp. 142–160. doi: 10.60084/hjas.v4i2.462.

Issue

Vol. 4 No. 2 (2026): September 2026

Section

Articles

License

Copyright (c) 2026 Teuku Rizky Noviandy, Mohsina Patwekar, Faheem Patwekar

Creative Commons License

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.