Deep generative molecular design reshapes drug discovery

Abstract

Recent advances and accomplishments of artificial intelligence (AI) and deep generative models have established their usefulness in medicinal applications, especially in drug discovery and development. To correctly apply AI, the developer and user face questions such as which protocols to consider, which factors to scrutinize, and how the deep generative models can integrate the relevant disciplines. This review summarizes classical and newly developed AI approaches, providing an updated and accessible guide to the broad computational drug discovery and development community. We introduce deep generative models from different standpoints and describe the theoretical frameworks for representing chemical and biological structures and their applications. We discuss the data and technical challenges and highlight future directions of multimodal deep generative models for accelerating drug discovery.

Figure 1

AI and deep generative model applications in the drug discovery pipeline Several successful applications of AI and deep generative models in various stage of the drug development pipeline: (A) AI-assistant target selection and validation, (B) molecular design, lead optimization, and chemical synthesis, (C) biological evaluation (in vitro and in vivo), clinical development, and post marketing surveillance, and (D) several successful preclinical and clinical molecules identified by AI and deep generative models. DDR1, discoidin domain receptor 1; DDR2, discoidin domain receptor tyrosine kinase 2; GSK3B, glycogen synthase kinase 3 beta; JNK3, c-Jun N-terminal kinase 3.

Figure 2

A diagram illustrating three molecular representation approaches Three molecular representation approaches include: (A) one-dimensional (1D) sequence-based representation; (B) graph-based representation; and (C) 3D representation for both small molecules and macromolecules (i.e., proteins). The value of contact map matrix is 1 if the distance is greater than a predetermined threshold, otherwise it is 0.

Figure 3

A diagram illustrating the theory framework of five deep generative models (A–E) in the drug discovery applications RNN, recurrent neural networks; VAE, variational autoencoder; GAN, generative adversarial networks; RL, reinforcement learning.

Figure 4

A proposed multimodal generative model in the drug discovery applications (A) A hybrid data model can fully capture diverse information during drug design, including chemical, drug-target interactions, drug-disease knowledge, and disease-relevant expression of target (protein/gene). (B) A multimodal generative model can consider various drug discovery pipeline components to increase likelihood of success of clinical trials. ADME-Tox, absorption, distribution, metabolism, and excretion-toxicity; IC₅₀, half-maximal inhibitory concentration.