De novo generation of hit-like molecules from gene expression signatures using artificial intelligence

Abstract

Finding new molecules with a desired biological activity is an extremely difficult task. In this context, artificial intelligence and generative models have been used for molecular de novo design and compound optimization. Herein, we report a generative model that bridges systems biology and molecular design, conditioning a generative adversarial network with transcriptomic data. By doing so, we can automatically design molecules that have a high probability to induce a desired transcriptomic profile. As long as the gene expression signature of the desired state is provided, this model is able to design active-like molecules for desired targets without any previous target annotation of the training compounds. Molecules designed by this model are more similar to active compounds than the ones identified by similarity of gene expression signatures. Overall, this method represents an alternative approach to bridge chemistry and biology in the long and difficult road of drug discovery.

De novo design and bioactivity prediction of SARS-CoV-2 main protease inhibitors using recurrent neural network-based transfer learning

“…In another study, Méndez-Lucio and coworkers trained a generative model using chemical (e..g, SMILES strings) and transcriptomic data for 978 genes, showing that the model could generate molecules that were structurally similar to their closest neighbors in the training set, while also introducing a range of modifications to the scaffolds. Concretely, starting with a benzene ring, the authors obtained structures with fused rings, different substitution patterns and also the replacement of carbons atoms to generate heterocycles [ 50 ]. A recent approach described by Arús-Pous et al was used to generate molecules starting from any scaffold; by exhaustively slicing acyclic bonds of the molecules on the training set the authors obtained a vast number of scaffolds and decorators data.…”

Deep learning and generative methods in cheminformatics and chemical biology: navigating small molecule space intelligently

“…What we would really like to have is the ability to generate the molecules themselves 'de novo' (e.g. [59,64,65,84,[92][93][94][95][96][97][98][99][100][101][102][103][104][105][106][107]), by learning what amounts to a joint distribution over all the variables (both inputs and outputs). To this end, a generative model seeks to simulate or recreate how the data are generated 'in the real world'.…”

Advances in De Novo Drug Design: From Conventional to Machine Learning Methods

“…Most of the approaches proposed to date are chemocentric, but new methodologies that bridge toxicogenomics and molecular design are starting to emerge. For example, Mendez-Lucio et al developed a DL model based on a GAN whose training was conditioned by gene expression data [ 138 ]. In a conditional generative model, target properties for each compound are incorporated into the training and generative phases, in addition to the compound chemical representation.…”