An explainable multi-parameter optimization approach for de novo drug design against proteins from central nervous system

Motivation: Aim of a successful drug design and development is to produce a drug which can inhibit the target protein and has a balanced physicochemical and toxicity profile. Traditionally, this is a multi-step process where different parameters such as activity, physicochemical and pharmacokinetic properties are optimized sequentially, which often leads to high attrition rate during later stages of drug design and development. Results: We have developed a deep learning-based de novo drug design method which can design novel small molecules by optimizing target-specificity as well as multiple parameters (including late stage parameters) in a single step. The model predictions were explained in two ways to remediate the black box nature of deep learning models: (1) the contribution of each parameter during multi-parameter optimization was computed using an adaptation of the SHAP algorithm and (2) an explainable predictive model was used to identify functional groups responsible for the property being optimized. The proposed method was validated against the human 5-hydroxy tryptamine receptor 1B (5-HT1B), a protein from the central nervous system (CNS). Various physicochemical properties specific to CNS drugs were considered along with the target specificity and blood-brain barrier (BBB), which acts as an additional challenge for CNS drug delivery. The contribution of each parameter towards molecule design was identified. The optimized generative model was able to design similar and better inhibitors compared to known inhibitors of 5-HT1B. In addition, the functional groups of the generated small molecules that guide the BBB permeability predictive model were identified through feature attribution techniques.