A Very Large-Scale Bioactivity Comparison of Deep Learning and Multiple Machine Learning Algorithms for Drug Discovery

Machine learning methods are attracting considerable attention from the pharmaceutical industry for use in drug discovery and applications beyond. In recent studies we have applied multiple machine learning algorithms, modeling metrics and in some cases compared molecular descriptors to build models for individual targets or properties on a relatively small scale. Several research groups have used large numbers of datasets from public databases such as ChEMBL in order to evaluate machine learning methods of interest to them. The largest of these types of studies used on the order of 1400 datasets. We have now extracted well over 5000 datasets from CHEMBL for use with the ECFP6 fingerprint and comparison of our proprietary software Assay CentralTM with random forest, k-Nearest Neighbors, support vector classification, naïve Bayesian, AdaBoosted decision trees, and deep neural networks (3 levels). Model performance was assessed using an array of five-fold cross-validation metrics including area-under-the-curve, F1 score, Cohen’s kappa and Matthews correlation coefficient. Based on ranked normalized scores for the metrics or datasets all methods appeared comparable while the distance from the top indicated Assay CentralTM and support vector classification were comparable. Unlike prior studies which have placed considerable emphasis on deep neural networks (deep learning), no advantage was seen in this case where minimal tuning was performed of any of the methods. If anything, Assay CentralTM may have been at a slight advantage as the activity cutoff for each of the over 5000 datasets representing over 570,000 unique compounds was based on Assay CentralTMperformance, but support vector classification seems to be a strong competitor. We also apply Assay CentralTM to prospective predictions for PXR and hERG to further validate these models. This work currently appears to be the largest comparison of machine learning algorithms to date. Future studies will likely evaluate additional databases, descriptors and algorithms, as well as further refining methods for evaluating and comparing models.