Display options
Cite
Kpanou R, Osseni MA, Tossou P, et al. On the robustness of generalization of drug-drug interaction models. BMC Bioinformatics. 2021;22(1):477doi: 10.1186/s12859-021-04398-9.
Kpanou, R., Osseni, M. A., Tossou, P., Laviolette, F., & Corbeil, J. (2021). On the robustness of generalization of drug-drug interaction models. BMC bioinformatics, 22(1), 477. https://doi.org/10.1186/s12859-021-04398-9
Kpanou, Rogia, et al. "On the robustness of generalization of drug-drug interaction models." BMC bioinformatics vol. 22,1 (2021): 477. doi: https://doi.org/10.1186/s12859-021-04398-9
Kpanou R, Osseni MA, Tossou P, Laviolette F, Corbeil J. On the robustness of generalization of drug-drug interaction models. BMC Bioinformatics. 2021 Oct 04;22(1):477. doi: 10.1186/s12859-021-04398-9. PMID: 34607569; PMCID: PMC8489092.
Copy Download .nbib Format: NLM
Share it on

BMC Bioinformatics. 2021 Oct 04;22(1):477. doi: 10.1186/s12859-021-04398-9.

On the robustness of generalization of drug-drug interaction models.

BMC bioinformatics

Rogia Kpanou, Mazid Abiodoun Osseni, Prudencio Tossou, Francois Laviolette, Jacques Corbeil

Affiliations

  1. Computer Science and Software Engineering, Université Laval, 1065, av. de la Médecine, Quebec, CA, Canada. [email protected].
  2. InVivo AI, Mila - 180 Corporate Lab L, 6650, 01 Rue Saint-Urbain, Montreal, CA, H2S 3G9, Canada. [email protected].
  3. Computer Science and Software Engineering, Université Laval, 1065, av. de la Médecine, Quebec, CA, Canada.
  4. InVivo AI, Mila - 180 Corporate Lab L, 6650, 01 Rue Saint-Urbain, Montreal, CA, H2S 3G9, Canada.
  5. Department of Molecular Medicine, Université Laval, 1065, av. de la Médecine, Quebec, CA, Canada.

PMID: 34607569 PMCID: PMC8489092 DOI: 10.1186/s12859-021-04398-9

Abstract

BACKGROUND: Deep learning methods are a proven commodity in many fields and endeavors. One of these endeavors is predicting the presence of adverse drug-drug interactions (DDIs). The models generated can predict, with reasonable accuracy, the phenotypes arising from the drug interactions using their molecular structures. Nevertheless, this task requires improvement to be truly useful. Given the complexity of the predictive task, an extensive benchmarking on structure-based models for DDIs prediction was performed to evaluate their drawbacks and advantages.

RESULTS: We rigorously tested various structure-based models that predict drug interactions using different splitting strategies to simulate different real-world scenarios. In addition to the effects of different training and testing setups on the robustness and generalizability of the models, we then explore the contribution of traditional approaches such as multitask learning and data augmentation.

CONCLUSION: Structure-based models tend to generalize poorly to unseen drugs despite their ability to identify new DDIs among drugs seen during training accurately. Indeed, they efficiently propagate information between known drugs and could be valuable for discovering new DDIs in a database. However, these models will most probably fail when exposed to unknown drugs. While multitask learning does not help in our case to solve the problem, the use of data augmentation does at least mitigate it. Therefore, researchers must be cautious of the bias of the random evaluation scheme, especially if their goal is to discover new DDIs.

© 2021. The Author(s).

Keywords: Deep learning; Drug–drug interaction; Generalizability; Robustness; Side effects

References

  1. PLoS Comput Biol. 2011 Dec;7(12):e1002323 - PubMed
  2. Bioinformatics. 2021 Mar 15;: - PubMed
  3. J Cheminform. 2015 May 20;7:20 - PubMed
  4. Brief Bioinform. 2018 Sep 28;19(5):863-877 - PubMed
  5. Perspect Clin Res. 2017 Oct-Dec;8(4):180-186 - PubMed
  6. Proc Natl Acad Sci U S A. 2018 May 1;115(18):E4304-E4311 - PubMed
  7. Mol Biosyst. 2016 Feb;12(2):614-23 - PubMed
  8. Bioinformatics. 2020 Aug 1;36(15):4316-4322 - PubMed
  9. PLoS One. 2019 Aug 1;14(8):e0219796 - PubMed
  10. J Am Med Inform Assoc. 2014 Oct;21(e2):e278-86 - PubMed
  11. Nat Biotechnol. 2014 Dec;32(12):1213-22 - PubMed
  12. J Cheminform. 2015 Aug 01;7:36 - PubMed
  13. J Am Coll Cardiol. 2016 Oct 18;68(16):1756-1764 - PubMed
  14. Science. 2006 Sep 29;313(5795):1929-35 - PubMed
  15. BMC Bioinformatics. 2021 Jun 11;22(1):318 - PubMed
  16. BMC Bioinformatics. 2017 Oct 10;18(1):445 - PubMed
  17. iScience. 2019 May 31;15:291-306 - PubMed
  18. PLoS Comput Biol. 2012;8(8):e1002614 - PubMed
  19. J Biomed Inform. 2018 Oct;86:15-24 - PubMed
  20. Sci Transl Med. 2012 Mar 14;4(125):125ra31 - PubMed
  21. NPJ Syst Biol Appl. 2017 Aug 25;3:23 - PubMed
  22. J Cheminform. 2017 Mar 7;9:16 - PubMed
  23. Brief Bioinform. 2016 Jan;17(1):2-12 - PubMed
  24. Sci Rep. 2019 Sep 20;9(1):13645 - PubMed
  25. Nucleic Acids Res. 2004 Jan 1;32(Database issue):D267-70 - PubMed
  26. Pac Symp Biocomput. 2012;:410-21 - PubMed
  27. Methods. 2020 Jul 1;179:37-46 - PubMed
  28. Brief Bioinform. 2021 May 05;: - PubMed
  29. BMC Bioinformatics. 2020 Sep 24;21(1):419 - PubMed
  30. BMC Bioinformatics. 2017 Oct 16;18(Suppl 12):409 - PubMed
  31. J Cheminform. 2015 Feb 26;7:7 - PubMed
  32. Bioinformatics. 2014 Jun 15;30(12):i228-36 - PubMed
  33. Nat Commun. 2015 Sep 28;6:8481 - PubMed
  34. Bioinformatics. 2021 Mar 26;: - PubMed
  35. Sci Rep. 2014 Nov 24;4:7160 - PubMed
  36. Sci Rep. 2015 Jul 21;5:12339 - PubMed
  37. J Cheminform. 2019 Nov 21;11(1):71 - PubMed
  38. Bioinformatics. 2019 Dec 15;35(24):5249-5256 - PubMed
  39. Pac Symp Biocomput. 2016;21:81-92 - PubMed
  40. BMC Bioinformatics. 2019 Aug 6;20(1):415 - PubMed
  41. J Chem Inf Model. 2010 May 24;50(5):742-54 - PubMed
  42. Bioinformatics. 2018 Jul 1;34(13):i457-i466 - PubMed
  43. PLoS Comput Biol. 2016 Jul 14;12(7):e1004975 - PubMed
  44. PLoS One. 2015 Mar 04;10(3):e0118432 - PubMed

Substances

MeSH terms

Publication Types