Toward efficient generation, correction, and properties control of unique drug‐like structures

Abstract

Efficient design and screening of the novel molecules is a major challenge in drug and material design. This paper focuses on a multi‐stage pipeline, in which several deep neural network models are combined to map discrete molecular representations into continuous vector space to later generate from it new molecular structures with desired properties. Here, the Attention‐based Sequence‐to‐Sequence model is added to “spellcheck” and correct generated structures, while the oversampling in the continuous space allows generating candidate structures with desired distribution for properties and molecular descriptors, even for a small reference datasets. We further use computer simulation to validate the desired properties in the numerical experiment. With the focus on the drug design, such a pipeline allows generating novel structures with a control of Synthetic Accessibility Score and a series of metrics that assess the drug‐likeliness. Our code is available at https://github.com/SoftServeInc/novel-molecule-generation.

Genetic Algorithms and Machine Learning for Predicting Surface Composition, Structure, and Chemistry: A Historical Perspective and Assessment

“…This method allowed pairing of the latent space descriptors produced by the VAE with a predictive NN to screen molecules in the database for desirable properties and generate new molecules with similar properties or with similarity to a particular species. This work has become popular as a technique for further molecular discovery. − VAE began to flourish among the general materials science community in 2019 with Stein et al predicting optical spectra from images of materials . Further work uses VAE for predicting materials synthesis, materials design, , and generating 3D printed material structures …”

Advancing molecular graphs with descriptors for the prediction of chemical reaction yields

“…Such ML models as variational autoencoders, 1,2 generative adversarial networks, 3,4 and recurrent neural networks 5,6 demonstrated promising de novo drug design capabilities. Recent advances in deep reinforcement learning (DRL) and Transformer‐like models took the molecule generation technology to a qualitatively next level 7–10 . DRL exhibited outstanding performance in optimization of chemical reaction conditions 11 and organic synthesis planning 12,13 …”

“…Recent advances in deep reinforcement learning (DRL) and Transformer-like models took the molecule generation technology to a qualitatively next level. [7][8][9][10] DRL exhibited outstanding performance in optimization of chemical reaction conditions 11 and organic synthesis planning. 12,13 This work assessed the opportunities and limitations of several ML approaches to predict an actual yield of organic reactions of multiple types.…”

XGBoost‐based intelligence yield prediction and reaction factors analysis of amination reaction

“…In recent years, machine learning algorithms have become an integral part of scientific inquiry in many disciplines, and have also brought new opportunities for the development of chemistry to realize the prediction of catalyst activation performance, 7–9 chemical reaction performance, 10–15 compound property prediction, 16–20 drug research and development, 21,22 auxiliary high‐performance materials design, 23 auxiliary inverse synthesis analysis, 24–26 and screening target compound 27 . Today, the information in a chemical system can be computed, screened, or encoded to form descriptor data for machine learning methods, and to help researchers conduct reasonable analysis and prediction and accelerate the research and development process of chemical research.…”