Improving machine learning in early drug discovery

  • Claus Bendtsen
  • Andrea Degasperi
  • Ernst Ahlberg
  • Lars Carlsson


The high cost for new medicines is hindering their development and machine learning is therefore being used to avoid carrying out physical experiments. Here, we present a comparison between three different machine learning approaches in a classification setting where learning and prediction follow a teaching schedule to mimic the drug discovery process. The approaches are standard SVM classification, SVM based multi-kernel classification and SVM classification based on learning using privileged information. Our two main conclusions are derived using experimental in-vitro data and compound structure descriptors. The in-vitro data is assumed to i) be completely absent in the standard SVM setting, ii) be available at all times when applying multi-kernel learning, or iii) be available as privileged information during training only. The structure descriptors are always available. One conclusion is that multi-kernel learning has higher odds than standard SVM in producing higher accuracy. The second is that learning using privileged information does not have higher odds than the standard SVM, although it may improve accuracy when the training sets are small.


Support vector machine SVM+ Privileged information Multi-kernel learning Human microsome clearance 

Mathematics Subject Classification (2010)

68T01 68Q32 92E10 


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.



AD is supported by the Science Foundation Ireland Industry Fellowship No. 15/IFA/2925.


  1. 1.
    Agresti, A.: Categorical Data Analysis. John Wiley & Sons, Inc., Hooken (2001)zbMATHGoogle Scholar
  2. 2.
    Arrowsmith, J., Miller, P.: Trial Watch: Phase II and Phase III attrition rates 2011–2012. Nat. Publ. Group 12(8), 569–569 (2013)Google Scholar
  3. 3.
    Ballard, P., Brassil, P., Bui, K.H., Dolgos, H., Petersson, C., Tunek, A., Webborn, P.J.H.: The right compound in the right assay at the right time: an integrated discovery DMPK strategy. Drug Metab. Rev. 44(3), 224–252 (2012)CrossRefGoogle Scholar
  4. 4.
    Chang, C.C., Lin, C.J.: LIBSVM: A library for support vector machines. ACM Trans. Intell. Syst. Technol. 2, 27:1–27:27 (2011). Software available at CrossRefGoogle Scholar
  5. 5.
    Cook, D., Brown, D., Alexander, R., March, R., Morgan, P., Satterthwaite, G., Pangalos, M.N.: Lessons learned from the fate of AstraZeneca’s drug pipeline: a five-dimensional framework. Nat. Publ. Group 13(6), 419–431 (2014)Google Scholar
  6. 6.
    Costello, J.C., Heiser, L.M., Georgii, E., Nen, M.G.O., Menden, M.P., Wang, N.J., Bansal, M., Ammadud din, M., Hintsanen, P., Khan, S.A., Mpindi, J.P., Kallioniemi, O., Honkela, A., Aittokallio, T., Wennerberg, K., Collins, J.J., Gallahan, D., Singer, D., Saez-Rodriguez, J., Kaski, S., Gray, J.W., Stolovitzky, G.: A community effort to assess and improve drug sensitivity prediction algorithms. Nat. Biotechnol. 32, 1202–1212 (2014)Google Scholar
  7. 7.
    DiMasi, J.A.: Cost of Developing a New Drug. Tech. Rep. R&D Cost Study Briefing, Tufts Center for the Study of Drug Development, Boston, MA (2014)Google Scholar
  8. 8.
    Eckert, H., Bajorath, J.: Molecular similarity analysis in virtual screening: foundations, limitations and novel approaches. Drug Discov. Today 12(5-6), 225–233 (2007)CrossRefGoogle Scholar
  9. 9.
    Eklund, M., Norinder, U., Boyer, S., Carlsson, L.: The application of conformal prediction to the drug discovery process. Annals of Mathematics and Artificial Intelligence pp. 1–16. doi: 10.1007/s10472-013-9378-2 (2013)
  10. 10.
    Eklund, M., Norinder, U., Boyer, S., Carlsson, L.: The application of conformal prediction to the drug discovery process. Ann. Math. Artif. Intell. 74(1), 117–132 (2015)MathSciNetCrossRefzbMATHGoogle Scholar
  11. 11.
    Faulon, J.L., Churchwell, C.J., Visco, D.P.: The Signature Molecular Descriptor. 2. Enumerating Molecules from Their Extended Valence Sequences. J. Chem. Inf. Comput. Sci. 43(3), 721–734 (2003)CrossRefGoogle Scholar
  12. 12.
    Faulon, J.L., Visco, D.P., Pophale, R.S.: The Signature Molecular Descriptor. 1. Using Extended Valence Sequences in QSAR and QSPR Studies. J. Chem. Inf. Comput. Sci. 43(3), 707–720 (2003)CrossRefGoogle Scholar
  13. 13.
    Gönen, M.: Alpaydın, E.: Multiple kernel learning algorithms. J. Mach. Learn. Res. 12(Jul), 2211–2268 (2011)MathSciNetzbMATHGoogle Scholar
  14. 14.
    Helal, K.Y., Maciejewski, M., Gregori-Puigjané, E., Glick, M., Wassermann, A.M.: Public Domain HTS Fingerprints: Design and Evaluation of Compound Bioactivity Profiles from PubChem’s Bioassay Repository. Journal of Chemical Information and Modeling p. acs.jcim.5b00498. doi: 10.1021/acs.jcim.5b00498 (2016)
  15. 15.
    Herper, M.: The Truly Staggering Cost Of Inventing New Drugs. Forbes (2012)Google Scholar
  16. 16.
    Lapin, M., Hein, M., Schiele, B.: Learning using privileged information: SVM+ and weighted SVM. Neural Netw. 53, 95–108 (2014)CrossRefzbMATHGoogle Scholar
  17. 17.
    Li, W., Dai, D., Tan, M., Xu, D., Van Gool, L.: Fast algorithms for linear and kernel SVM+ Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2258–2266 (2016)Google Scholar
  18. 18.
    Liang, L., Cherkassky, V.: Connection between svm+ and multi-task learning 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence), pp. 2048–2054. IEEE (2008)Google Scholar
  19. 19.
    Liu, R., Schyman, P., Wallqvist, A.: Critically assessing the predictive power of qsar models for human liver microsomal stability. J. Chem. Inf. Model. 55(8), 1566–1575 (2015)CrossRefGoogle Scholar
  20. 20.
    Mayr, A., Klambauer, G., Unterthiner, T., Hochreiter, S.: DeepTox: Toxicity Prediction using Deep Learning. Front. Environ. Sci. 3, 24–15 (2016)CrossRefGoogle Scholar
  21. 21.
    Pasupa, K., Hussain, Z., Shawe-Taylor, J., Willett, P.: Drug screening with elastic-net multiple kernel learning 13th IEEE International Conference on BioInformatics and BioEngineering, pp 1–5 (2013). doi: 10.1109/BIBE.2013.6701529
  22. 22.
    Paul, S.M., Mytelka, D.S., Dunwiddie, C.T., Persinger, C.C., Munos, B.H., Lindborg, S.R., Schacht, A.L.: How to improve R&D productivity: the pharmaceutical industry’s grand challenge. Nature Reviews Drug Discovery 1–12 (2010)Google Scholar
  23. 23.
    Pechyony, D., Izmailov, R., Vashist, A., Vapnik, V.: Smo-style algorithms for learning using privileged information DMIN, pp. 235–241 (2010)Google Scholar
  24. 24.
    Pechyony, D., Vapnik, V.: Fast optimization algorithms for solving svm+. Stat. Learning and Data Science 1 (2011)Google Scholar
  25. 25.
    Peck, R.W., Lendrem, D.W., Grant, I., Lendrem, B.C., Isaacs, J.D.: Why is it hard to terminate failing projects in pharmaceutical R&D?. Nature Publishing Group, 1–2 (2015)Google Scholar
  26. 26.
    Petrone, P.M., Simms, B., Nigsch, F., Lounkine, E., Kutchukian, P., Cornett, A., Deng, Z., Davies, J.W., Jenkins, J.L., Glick, M.: Rethinking molecular similarity: Comparing compounds on the basis of biological activity. ACS Chem. Biol. 7(8), 1399–1409 (2012). doi: 10.1021/cb3001028 CrossRefGoogle Scholar
  27. 27.
    Ribeiro, B., Silva, C., Chen, N., Vieira, A., das Neves, J.C.: Enhanced default risk models with SVM+. Expert Syst. Appl. 39(11), 10,140–10,152 (2012)CrossRefGoogle Scholar
  28. 28.
    Riniker, S., Wang, Y., Jenkins, J.L., Landrum, G.A.: Using information from historical high-throughput screens to predict active compounds. doi: 10.1021/ci500190p (2014)
  29. 29.
    Scannell, J.W., Bosley, J.: When Quality Beats Quantity: Decision Theory, Drug Discovery, and the Reproducibility Crisis. PLoS ONE 11(2), e0147,215–21 (2016)CrossRefGoogle Scholar
  30. 30.
    Serra-Toro, C., Traver, V.J., Pla, F.: Exploring some practical issues of SVM+: Is really privileged information that helps Pattern Recogn. Lett. 42, 40–46 (2014)CrossRefGoogle Scholar
  31. 31.
    Steinbeck, C., Han, Y., Kuhn, S., Horlacher, O., Luttmann, E., Willighagen, E.: The chemistry development kit (cdk) an open-source java library for chemo- and bioinformatics. J. Chem. Inf. Comput. Sci. 43(2), 493–500 (2003). doi: 10.1021/ci025584y PMID: 12653513CrossRefGoogle Scholar
  32. 32.
    Vapnik, V.: Learning Using Privileged Information: Similarity Control and Knowledge Transfer (2015)Google Scholar
  33. 33.
    Vapnik, V., Vashist, A.: A new learning paradigm: Learning using privileged information. Neural Netw. 22(5), 544–557 (2009)CrossRefzbMATHGoogle Scholar
  34. 34.
    Vovk, V., Shafer, G., Gammerman, A.: Algorithmic learning in a random world. Springer, New York (2005)zbMATHGoogle Scholar
  35. 35.
    Wang, Z., Ji, Q.: Classifier learning with hidden information Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp 4969–4977 (2015)Google Scholar
  36. 36.
    Waring, M.J., Arrowsmith, J., Leach, A.R., Leeson, P.D., Mandrell, S., Owen, R.M., Pairaudeau, G., Pennie, W.D., Pickett, S.D., Wang, J., Wallace, O., Weir, A.: An analysis of the attrition of drug candidates from four major pharmaceutical companies. Nat. Publ. Group 14(7), 475–486 (2015)Google Scholar
  37. 37.
    Woolf, B.: On estimating the relation between blood group and disease. Ann. Human Genet. 19, 251–253 (1955)CrossRefGoogle Scholar
  38. 38.
    Xu, X., Zhou, J.T., Tsang, I., Qin, Z., Goh, R.S.M., Liu, Y.: Simple and efficient learning using privileged information BeyondLabeler: Human is More Than a Labeler. Workshop of the 25th International Joint Conference on Artificial Intelligence (IJCAI-16), New York City, USA. arXiv:1604.01518(2016)
  39. 39.
    Yau, E., Petersson, C., Dolgos, H., Peters, S.A.: A comparative evaluation of models to predict human intestinal metabolism from nonclinical data. Biopharmaceutics & Drug Disposition (2017)Google Scholar

Copyright information

© Springer International Publishing Switzerland 2017

Authors and Affiliations

  1. 1.AstraZeneca, Innovative Medicines & Early DevelopmentQuantitative Biology, Discovery SciencesCambridgeUK
  2. 2.University College Dublin Systems Biology IrelandBelfiledRepublic of Ireland
  3. 3.AstraZeneca, Innovative Medicines & Early DevelopmentPredictive Compound ADME & Safety, Drug Safety & MetabolismMölndalSweden
  4. 4.AstraZeneca, Innovative Medicines & Early DevelopmentQuantitative Biology, Discovery SciencesMölndalSweden

Personalised recommendations