Molecular machine learning with conformer ensembles DOI Creative Commons
Simon Axelrod, Rafael Gómez‐Bombarelli

Machine Learning Science and Technology, Journal Year: 2023, Volume and Issue: 4(3), P. 035025 - 035025

Published: Aug. 11, 2023

Abstract Virtual screening can accelerate drug discovery by identifying promising candidates for experimental evaluation. Machine learning is a powerful method screening, as it learn complex structure–property relationships from data and make rapid predictions over virtual libraries. Molecules inherently exist three-dimensional ensemble their biological action typically occurs through supramolecular recognition. However, most deep approaches to molecular property prediction use 2D graph representation input, in some cases single 3D conformation. Here we investigate how the information of multiple conformers, traditionally known 4D cheminformatics community, improve models. We introduce models that expand upon key architectures such ChemProp SchNet, adding elements multiple-conformer inputs conformer attention. then benchmark performance trade-offs these on 2D, representations activity using large training set geometrically resolved molecules. The new perform significantly better than models, but often just strong with many. also find interpretable attention weights each conformer.

Language: Английский

A critical overview of computational approaches employed for COVID-19 drug discovery DOI Creative Commons
Eugene Muratov, Rommie E. Amaro, Carolina Horta Andrade

et al.

Chemical Society Reviews, Journal Year: 2021, Volume and Issue: 50(16), P. 9121 - 9151

Published: Jan. 1, 2021

COVID-19 has resulted in huge numbers of infections and deaths worldwide brought the most severe disruptions to societies economies since Great Depression. Massive experimental computational research effort understand characterize disease rapidly develop diagnostics, vaccines, drugs emerged response this devastating pandemic more than 130 000 COVID-19-related papers have been published peer-reviewed journals or deposited preprint servers. Much focused on discovery novel drug candidates repurposing existing against COVID-19, many such projects either exclusively computer-aided studies. Herein, we provide an expert overview key methods their applications for small-molecule therapeutics that reported literature. We further outline that, after first year pandemic, it appears not produced rapid global solutions. However, several known used clinic cure patients, a few repurposed continue be considered clinical trials, along with candidates. posit truly impactful tools must deliver actionable, experimentally testable hypotheses enabling combinations, open science sharing results are critical accelerate development novel, much needed COVID-19.

Language: Английский

Citations

181
Deep generative molecular design reshapes drug discovery DOI Creative Commons

Xiangxiang Zeng,

Fei Wang, Yuan Luo

et al.

Cell Reports Medicine, Journal Year: 2022, Volume and Issue: 3(12), P. 100794 - 100794

Published: Oct. 27, 2022

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

Language: Английский

Citations

145
Uni-Mol: A Universal 3D Molecular Representation Learning Framework DOI Creative Commons

Gengmo Zhou,

Zhifeng Gao,

Qiankun Ding

et al.

Published: March 7, 2023

Molecular representation learning (MRL) has gained tremendous attention due to its critical role in from limited supervised data for applications like drug design. In most MRL methods, molecules are treated as 1D sequential tokens or 2D topology graphs, limiting their ability incorporate 3D information downstream tasks and, particular, making it almost impossible geometry prediction/generation. this paper, we propose a universal framework, called Uni-Mol, that significantly enlarges the and application scope of schemes. Uni-Mol contains two pretrained models with same SE(3) Transformer architecture: molecular model by 209M conformations; pocket 3M candidate protein data. Besides, several finetuning strategies apply various tasks. By properly incorporating information, outperforms SOTA 14/15 property prediction Moreover, achieves superior performance spatial tasks, including protein-ligand binding pose prediction, conformation generation, etc. The code, model, made publicly available at https://github.com/dptech-corp/Uni-Mol.

Language: Английский

Citations

109
CREST—A program for the exploration of low-energy molecular chemical space DOI Creative Commons
Philipp Pracht, Stefan Grimme, Christoph Bannwarth

et al.

The Journal of Chemical Physics, Journal Year: 2024, Volume and Issue: 160(11)

Published: March 21, 2024

Conformer–rotamer sampling tool (CREST) is an open-source program for the efficient and automated exploration of molecular chemical space. Originally developed in Pracht et al. [Phys. Chem. Phys. 22, 7169 (2020)] as driver calculations at extended tight-binding level (xTB), it offers a variety molecular- metadynamics simulations, geometry optimization, structure analysis capabilities. Implemented algorithms include procedures conformational sampling, explicit solvation studies, calculation absolute entropy, identification protonation deprotonation sites. Calculations are set up to run concurrently, providing single-node parallelization. CREST designed require minimal user input comes with implementation GFNn-xTB Hamiltonians GFN-FF force-field. Furthermore, interfaces any quantum chemistry force-field software can easily be created. In this article, we present recent developments code show selection applications most important features program. An novelty refactored backend, which provides significant speed-up small or medium-sized drug molecules allows more sophisticated setups, example, mechanics/molecular mechanics minimum energy crossing point calculations.

Language: Английский

Citations

98
QMugs, quantum mechanical properties of drug-like molecules DOI Creative Commons
Clemens Isert, Kenneth Atz, José Jiménez-Luna

et al.

Scientific Data, Journal Year: 2022, Volume and Issue: 9(1)

Published: June 7, 2022

Abstract Machine learning approaches in drug discovery, as well other areas of the chemical sciences, benefit from curated datasets physical molecular properties. However, there currently is a lack data collections featuring large bioactive molecules alongside first-principle quantum information. The open-access QMugs (Quantum-Mechanical Properties Drug-like Molecules) dataset fills this void. collection comprises mechanical properties more than 665 k biologically and pharmacologically relevant extracted ChEMBL database, totaling ~2 M conformers. contains optimized geometries thermodynamic obtained via semi-empirical method GFN2-xTB. Atomic are provided on both GFN2-xTB density-functional levels theory (DFT, ω B97X-D/def2-SVP). features significantly larger size previously-reported their respective wave functions, including DFT density orbital matrices. This intended to facilitate development models that learn different while also providing insight into corresponding relationships between structure biological activity.

Language: Английский

Citations

85
Multi-modal molecule structure–text model for text-based retrieval and editing DOI
Shengchao Liu, Weili Nie, Chengpeng Wang

et al.

Nature Machine Intelligence, Journal Year: 2023, Volume and Issue: 5(12), P. 1447 - 1457

Published: Dec. 18, 2023

Language: Английский

Citations

62
New avenues in artificial-intelligence-assisted drug discovery DOI Creative Commons
Carmen Cerchia, Antonio Lavecchia

Drug Discovery Today, Journal Year: 2023, Volume and Issue: 28(4), P. 103516 - 103516

Published: Feb. 2, 2023

Over the past decade, amount of biomedical data available has grown at unprecedented rates. Increased automation technology and larger volumes have encouraged use machine learning (ML) or artificial intelligence (AI) techniques for mining such extracting useful patterns. Because identification chemical entities with desired biological activity is a crucial task in drug discovery, AI technologies potential to accelerate this process support decision making. In addition, advent deep (DL) shown great promise addressing diverse problems as de novo molecular design. Herein, we will appraise current state-of-the-art AI-assisted discussing recent applications covering generative models structure generation, scoring functions improve binding affinity pose prediction, dynamics assist parametrization, featurization generalization tasks. Finally, discuss hurdles strategies overcome them, well future directions.

Language: Английский

Citations

59
High-throughput property-driven generative design of functional organic molecules DOI
Julia Westermayr, Joe Gilkes,

Rhyan Barrett

et al.

Nature Computational Science, Journal Year: 2023, Volume and Issue: 3(2), P. 139 - 148

Published: Feb. 6, 2023

Language: Английский

Citations

50
Graph neural networks DOI
Gabriele Corso, H. Stärk,

Stefanie Jegelka

et al.

Nature Reviews Methods Primers, Journal Year: 2024, Volume and Issue: 4(1)

Published: March 7, 2024

Language: Английский

Citations

49
ResGen is a pocket-aware 3D molecular generation model based on parallel multiscale modelling DOI

Odin Zhang,

Jintu Zhang,

Jieyu Jin

et al.

Nature Machine Intelligence, Journal Year: 2023, Volume and Issue: 5(9), P. 1020 - 1030

Published: Sept. 7, 2023

Language: Английский

Citations

44