Modeling Physico-Chemical ADMET Endpoints with Multitask Graph Convolutional Networks

Abstract

Simple physico-chemical properties, like logD, solubility, or melting point, can reveal a great deal about how a compound under development might later behave. These data are typically measured for most compounds in drug discovery projects in a medium throughput fashion. Collecting and assembling all the Bayer in-house data related to these properties allowed us to apply powerful machine learning techniques to predict the outcome of those assays for new compounds. In this paper, we report our finding that, especially for predicting physicochemical ADMET endpoints, a multitask graph convolutional approach appears a highly competitive choice. For seven endpoints of interest, we compared the performance of that approach to fully connected neural networks and different single task models. The new model shows increased predictive performance compared to previous modeling methods and will allow early prioritization of compounds even before they are synthesized. In addition, our model follows the generalized solubility equation without being explicitly trained under this constraint.

Molecular Property Prediction Based on a Multichannel Substructure Graph

“…The prediction of molecular properties plays a vital role in drug discovery [1], [2]. Traditional methods such as biochemical experiments are always time-consuming and expensive.…”

Artificial intelligence in drug discovery: recent advances and future perspectives

“…The idea of multi-output QSAR modeling, aiming to relate a set of predefined chemical descriptors to observable endpoints, had been explored before the rise of popularity of deep learning approaches [37][38][39][40][41][42]. Despite the promise of multitask learning, to date, only modest performance improvements over single-task models have been reported [43][44][45][46].…”

Neuraldecipher – reverse-engineering extended-connectivity fingerprints (ECFPs) to their molecular structures

“…Because drug candidates with high binding affinity can still fail in later phases of clinical trials due to poor pharmacokinetic and toxicological (ADMET) proles, modeling ADMET endpoints such as solubility or melting point, is nowadays also considered in in silico de novo drug design at early stages. 5 Securely exchanging chemical data without revealing the molecular structure is especially nowadays of great importance, as sharing data such as ngerprints and/or measured endpoints between research groups within academia or private sectors through collaborations is oen accomplished to improve drug discovery.…”

“…there are more unique ECFP samples. The barplot displays the counts for the degeneracies[2,3,4,5]. Degeneracies larger than 6 are not displayed, since the frequency that 6 different structures map to the same ECFP is small.…”