Kinome-wide activity classification of small molecules by deep learning

Abstract

Deep learning is a machine learning technique that attempts to model high-level abstractions in data by utilizing a graph composed of multiple processing layers that experience various linear and non-linear transformations. This technique has been shown to perform well for applications in drug discovery, utilizing structural features of small molecules to predict activity. However, the application of deep learning to discriminating features of kinase inhibitors has not been well explored. Small molecule kinase inhibitors are an important class of anti-cancer agents and have demonstrated impressive clinical efficacy in several different diseases. However, resistance is often observed mediated by adaptive Kinome reprogramming or subpopulation diversity.Therefore, polypharmacology and combination therapies offer potential therapeutic strategies for patients with resistant disease. Their development would benefit from more comprehensive and dense knowledge of small-molecule inhibition across the human Kinome. Because such data is not publicly available, we evaluated multiple machine learning methods to predict small molecule inhibition of 342 kinases using over 650K aggregated bioactivity annotations for over 300K small molecules curated from ChEMBL and the Kinase Knowledge Base (KKB). Our results demonstrated that multi-task deep neural networks outperform classical single-task methods, offering potential towards predicting activity profiles and filling gaps in the available data. ANCILLARY INFORMATIONSupporting Information. Author ContributionsBKA. developed code, performed computational modeling and data analysis; data curation and data integration; BKA and SCS performed data curation and integration; SCS devised the project; SCS and NA advised the project; BKA and SCS wrote the manuscript.

Machine and deep learning approaches for cancer drug repurposing

“…Ligand-based deep learning approaches have also been of benefit. In one such model, Allen et al created a multi-task deep neural network (MTDNN) models to predict kinase activity of small molecules across the human kinome (31). Briefly, comprehensive kinase bioactivity data was obtained from ChEMBL and the commercial database Kinase Knowledgebase KKB (32).…”

“…In general, more data incorporated into the training set generally results in better model performance (less overfitting), especially for deep neural networks. For example, Allen et al found that for predicting drug-kinase activity using deep learning, model performance was significantly less when the number of active compounds for a kinase was less than 500 and further degraded when there was less than 50 (31). This is with respect to single-task learning (i.e.…”