Multi-Objective and Structure-Aware Generative Drug Design (MOSA).

Current molecular design tools are mostly uni-objective by focusing on the optimization of one primary property using ligand-based models. In this project, we aim to develop AI-based molecular design tools for multi-objective molecular design that are aware of 3D structures of the receptor and ligand candidates which will reduce attrition due to unmet secondary objectives. The enrichment with structural data will also include injecting more physical knowledge in the interaction of protein-ligand designs by utilizing conformational sampling and model enrichment by learning from the quantum mechanical calculation of the molecular structures. The project will be organized into three work packages: 1) The first work package addresses tasks related to Integrating target structural information in predictive and generative models - 3D models: Introducing a new 3D DL model based on structural data of the ligand-receptor complex. This model can also be fine-tuned in downstream tasks for predicting structure related properties like affinity and IC50. - 4D models: Integration of the conformational flexibility into the model from the previous task. - Structure-based generative model: by using the latent space of the models (3D or 4D) to design new molecules with new novel interactions with the receptor 2) The second work package tackles tasks related to multi-objective molecule optimization - Introducing new shape/conformational energetic penalty/pharmacophoric scoring functions which can explicitly be implemented in an automatic differentiable framework to enable end-to-end the optimization process using the powerful DL optimizers. - Predictive2generative: use direct discrete gradient-based optimization or gradient-based distributional optimization to enable using predictive models for molecular optimization - Supervised molecular generator using controllable decoding models or via conditional generator 3) The third work package is related to Integrating the delivered models for the WPs 1&2 in a unified framework and applying above outcomes on real world drug design business applications across all therapeutic areas within Janssen.

Keywords:THEORETICAL STUDY, MOLECULAR DYNAMICS

Disciplines:Drug discovery and development not elsewhere classified

Project type:Collaboration project