Generative Adversarial Networks for De Novo Molecular Design

Abstract

In the chemical industry, the generation of novel molecular structures with beneficial pharmacological and physicochemical properties in de novo molecular design is a critical problem. The advent of deep learning and neural generative models has recently enabled significant achievements in constructing molecular design models in de novo design. Consequently, studies on new generative models continue to generate molecules that exhibit more useful chemical properties. In this study, we propose a method for de novo design that utilizes generative adversarial networks based on reinforcement learning for realistic molecule generation. This method learns to reproduce the training data distribution of simplified molecular-input line-entry system strings. The proposed method is demonstrated to effectively generate novel molecular structures from five benchmark results using a real-world public dataset, ChEMBL. The code is available at https://github.com/ dudwojae/SMILES-MaskGAN.

68 million natural product-like compound database generated via molecular language processing

“…Compared to manually curated natural product libraries, deep generative models can transcend the boundaries of human imagination-dependent molecular design to significantly expand chemical search space by orders of magnitude while concurrently reducing financial and resource requirements. 17,18 Some examples of deep generative architectures that have been employed for molecular design include variational autoencoders (VAE), 19,20 recurrent neural networks (RNN), [21][22][23] and generative adversarial networks (GAN), [24][25][26] each adopting a different strategy with their own strengths and weaknesses. 27 In this work, we have trained a RNN-based long short-term memory (LSTM) network architecture 21 on tokenized SMILES (Simplified Molecular Input Line Entry System) 28 from 325,535 (80%) out of the 406,919 known natural products in COCONUT, the collection of open natural products (https://coconut.naturalproducts.net/, accessed 1 Aug 2022).…”

Potent antibiotic design via guided search from antibacterial activity evaluations

“…In generation-based work, many models represent a molecule with the corresponding simplified molecular input line entry specification (SMILES) notation so that the molecule generation task is transformed into a sequence-to-sequence generation task ( Born et al , 2021 ; Krishnan et al , 2021 ; Lee et al , 2021 ; Olivecrona et al , 2017 ; Popova et al , 2018 ; Segler et al , 2018a ; Wang et al , 2021 ). However, this approach often requires large-scale pretraining, and grammatical errors will often lead to the generation of invalid SMILES, which cannot be converted into realistic molecular structures.…”

Generation of novel Diels-Alder reactions using a generative adversarial network

“…6,7 Lee et al applied the generative adversarial network (GAN) to de novo molecular design and demonstrated outstanding performance in the five distribution learning benchmarks in the GuacaMol framework. 8 The remarkable achievements of generative models in molecular generation inspired the chemists, both Bort et al and Wang et al obtained the reactions of their interest with reaction generative models. 9,10 Generative models are an important class of models in machine learning in which we can generate new data that is not included in the training dataset and have shown its enormous potential in the field of image, 11 text 12 and sound generation 13 in the past few years.…”