Generative Models as an Emerging Paradigm in the Chemical Sciences

Abstract

Traditional computational approaches to design chemical species are limited by the need to compute properties for a vast number of candidates, e.g., by discriminative modeling. Therefore, inverse design methods aim to start from the desired property and optimize a corresponding chemical structure. From a machine learning viewpoint, the inverse design problem can be addressed through so-called generative modeling. Mathematically, discriminative models are defined by learning the probability distribution function of properties given the molecular or material structure. In contrast, a generative model seeks to exploit the joint probability of a chemical species with target characteristics. The overarching idea of generative modeling is to implement a system that produces novel compounds that are expected to have a desired set of chemical features, effectively sidestepping issues found in the forward design process. In this contribution, we overview and critically analyze popular generative algorithms like generative adversarial networks, variational autoencoders, flow, and diffusion models. We highlight key differences between each of the models, provide insights into recent success stories, and discuss outstanding challenges for realizing generative modeling discovered solutions in chemical applications.

Alchemical analysis of FDA approved drugs

“…These tools should generally be applicable to analyze drug sets from multiple angles in the context of drug discovery. One specic case could be the analysis of analog series obtained from generative models 53,54 to help identify feasible transformations or single out scaffold hopping changes.…”

In Pursuit of the Exceptional: Research Directions for Machine Learning in Chemical and Materials Science

“…As it is impossible to exhaustively enumerate all of the possibilities in these problems, one must instead sample the possibilities, which corresponds to the task of generative ML; methods and applications of generative ML to chemical problems have very recently been reviewed in ref 155. We advocate that these methods be used to cast a wide net.…”

DrugEx: Deep Learning Models and Tools for Exploration of Drug-Like Chemical Space

“…Rapid technological improvements over the last decades have led to the rising popularity of advanced machine learning methods. − These developments have also greatly influenced the field of DNDD with state-of-the-art methods including population-based metaheuristics, recurrent neural networks (RNNs), generative adversarial networks, variational autoencoders, and transformers. − Moreover, concepts such as transfer, conditional, and reinforcement learning (RL) are often applied to generate molecules with desired properties.…”