ConfRank: Improving GFN-FF Conformer Ranking with Pairwise Training DOI
Christian Hölzer,

Rick Oerder,

Stefan Grimme

et al.

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 20, 2024

Conformer ranking is a crucial task for drug discovery, with methods generating conformers often based on molecular (meta)dynamics or sophisticated sampling techniques. These are constrained by the underlying force computation regarding runtime and energy accuracy, limiting their effectiveness large-scale screening applications. To address these limitations, we introduce ConfRank, machine learning-based approach that enhances conformer using pairwise training. We demonstrate its performance GFN-FF-generated ensembles, leveraging DimeNet++ architecture trained pairs of 159 760 uncharged organic compounds from GEOM data set r2SCAN-3c reference level. Instead predicting only single molecules, this captures relative differences between conformers, leading to significant improvement overall conformational ranking, outperforming GFN-FF GFN2-xTB. Thereby, RMSD difference two can be reduced 5.65 0.71 kcal mol–1 test set, allowing correctly identify up 81% all lowest lying (GFN-FF: 10%, GFN2-xTB: 47%). The ConfRank cost-effective, scalable deployment both CPU GPU, achieving accelerations 2 orders magnitude compared Out-of-sample investigations CREST-generated ensembles QM9 taken an extended GMTKN55 show promising results robustness approach. correlation coefficient such as Spearman improved 0.90 0.39, 0.84) reducing probability incorrect sign flip in comparison 32 7%. On subsets MAD (RMSD) could almost 62% (58%) average 30% (29%). Moreover, exemplary case study vancomycin shows similar performance, indicating applicability larger (bio)molecular structures. Furthermore, motivate usage training theoretical perspective, highlighting while lead decline sample prediction absolute energies ML models, it significantly performance. models used available at https://github.com/grimme-lab/confrank.

Language: Английский

The Physics-AI Dialogue in Drug Design DOI Creative Commons
Pablo Andrés Vargas-Rosales, Amedeo Caflisch

RSC Medicinal Chemistry, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 1, 2025

A long path has led from the determination of first protein structure in 1960 to recent breakthroughs science. Protein prediction and design methodologies based on machine learning (ML) have been recognized with 2024 Nobel prize Chemistry, but they would not possible without previous work input many domain scientists. Challenges remain application ML tools for structural ensembles their usage within software pipelines by crystallography or cryogenic electron microscopy. In drug discovery workflow, techniques are being used diverse areas such as scoring docked poses, generation molecular descriptors. As become more widespread, novel applications emerge which can profit large amounts data available. Nevertheless, it is essential balance potential advantages against environmental costs deployment decide if when best apply it. For hit lead optimization efficiently interpolate between compounds chemical series free energy calculations dynamics simulations seem be superior designing derivatives. Importantly, complementarity and/or synergism physics-based methods (e.g., force field-based simulation models) data-hungry growing strongly. Current evolved decades research. It now necessary biologists, physicists, computer scientists fully understand limitations ensure that exploited design.

Language: Английский

Citations

1
The next revolution in computational simulations: Harnessing AI and quantum computing in molecular dynamics DOI
Anna Lappala

Current Opinion in Structural Biology, Journal Year: 2024, Volume and Issue: 89, P. 102919 - 102919

Published: Sept. 21, 2024

Language: Английский

Citations

7
A multiscale molecular structural neural network for molecular property prediction DOI
Zhiwei Shi, Miao Ma, Hanyang Ning

et al.

Molecular Diversity, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 25, 2025

Language: Английский

Citations

0
ABFML: A problem-oriented package for rapidly creating, screening, and optimizing new machine learning force fields DOI
Xingze Geng,

Jianing Gu,

Gaowu Qin

et al.

The Journal of Chemical Physics, Journal Year: 2025, Volume and Issue: 162(5)

Published: Feb. 4, 2025

Machine Learning Force Fields (MLFFs) require ongoing improvement and innovation to effectively address challenges across various domains. Developing MLFF models typically involves extensive screening, tuning, iterative testing. However, existing packages based on a single mature descriptor or model are unsuitable for this process. Therefore, we developed package named ABFML, PyTorch, which aims promote by providing developers with rapid, efficient, user-friendly tool constructing, validating new force field models. Moreover, leveraging standardized module operations cutting-edge machine learning frameworks, can swiftly establish In addition, the platform seamlessly transition graphics processing unit environments, enabling accelerated calculations large-scale parallel simulations of molecular dynamics. contrast traditional from-scratch approaches development, ABFML significantly lowers barriers developing models, thereby expediting application within development

Language: Английский

Citations

0
Applications of machine learning in surfaces and interfaces DOI Open Access
Shaofeng Xu, Jing‐Yuan Wu, Ying Guo

et al.

Chemical Physics Reviews, Journal Year: 2025, Volume and Issue: 6(1)

Published: March 1, 2025

Surfaces and interfaces play key roles in chemical material science. Understanding physical processes at complex surfaces is a challenging task. Machine learning provides powerful tool to help analyze accelerate simulations. This comprehensive review affords an overview of the applications machine study systems materials. We categorize into following broad categories: solid–solid interface, solid–liquid liquid–liquid surface solid, liquid, three-phase interfaces. High-throughput screening, combined first-principles calculations, force field accelerated molecular dynamics simulations are used rational design such as all-solid-state batteries, solar cells, heterogeneous catalysis. detailed information on for

Language: Английский

Citations

0
Molecular Modelling in Bioactive Peptide Discovery and Characterisation DOI Creative Commons
Clement Agoni, Raúl Fernández-Díaz, Patrick Brendan Timmons

et al.

Biomolecules, Journal Year: 2025, Volume and Issue: 15(4), P. 524 - 524

Published: April 3, 2025

Molecular modelling is a vital tool in the discovery and characterisation of bioactive peptides, providing insights into their structural properties interactions with biological targets. Many models predicting peptide function or structure rely on intrinsic properties, including influence amino acid composition, sequence, chain length, which impact stability, folding, aggregation, target interaction. Homology predicts structures based known templates. Peptide-protein can be explored using molecular docking techniques, but there are challenges related to inherent flexibility addressed by more computationally intensive approaches that consider movement over time, called dynamics (MD). Virtual screening many usually against single target, enables rapid identification potential peptides from large libraries, typically approaches. The integration artificial intelligence (AI) has transformed leveraging amounts data. AlphaFold general protein prediction deep learning greatly improved predictions conformations interactions, addition estimates model accuracy at each residue guide interpretation. Peptide being further enhanced Protein Language Models (PLMs), deep-learning-derived statistical learn computer representations useful identify fundamental patterns proteins. Recent methodological developments discussed context canonical as well those modifications cyclisations. In designing therapeutics, main outstanding challenge for these methods incorporation diverse non-canonical acids

Language: Английский

Citations

0
Bridging the Computational Gap: Sliding Window Technique Meets GCNN for Enhanced Molecular Charge Predictions DOI Creative Commons
Vicente Domínguez-Arca

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: March 15, 2024

Abstract In the quest for advancing computational tools capable of accurately calculating, estimating, or predicting partial atomic charges in organic molecules, this work introduces a pioneering Machine Learning-based tool designed to transcend limitations traditional methods like DFT, Mulliken, and semi-empirical approaches such as MOPAC Gaussian. Recognizing crucial role molecular dynamics simulations studying solvation, protein interactions, substrate membrane permeability, we aim introduce that not only offers enhanced efficiency but also extends predictive capabilities molecules larger than those QM9 dataset, traditionally analyzed using Mulliken charges. Employing novel neural network architecture adept at learning graph properties and, by extension, characteristics study presents "sliding window" technique. This method segments into smaller, manageable substructures charge prediction, significantly reducing demands processing times. Our results highlight model's accuracy unseen from database its successful application resveratrol molecule, providing insights hydrogen-donating CH groups aromatic rings—a feature predicted existing CGenFF ATB supported literature. breakthrough alternative determining chemistry underscores potential convolutional networks discern features based on stoichiometry geometric configuration. Such advancements hint future possibility designing with desired sequences, promising transformative impact drug discovery.

Language: Английский

Citations

0
Future Opportunities for Systematic AI Support in Healthcare DOI Creative Commons

Markus Bertl,

Gunnar Piho, Dirk Draheim

et al.

Lecture notes in computer science, Journal Year: 2024, Volume and Issue: unknown, P. 203 - 224

Published: Oct. 30, 2024

Language: Английский

Citations

0
ConfRank: Improving GFN-FF Conformer Ranking with Pairwise Training DOI
Christian Hölzer,

Rick Oerder,

Stefan Grimme

et al.

Journal of Chemical Information and Modeling, Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 20, 2024

Conformer ranking is a crucial task for drug discovery, with methods generating conformers often based on molecular (meta)dynamics or sophisticated sampling techniques. These are constrained by the underlying force computation regarding runtime and energy accuracy, limiting their effectiveness large-scale screening applications. To address these limitations, we introduce ConfRank, machine learning-based approach that enhances conformer using pairwise training. We demonstrate its performance GFN-FF-generated ensembles, leveraging DimeNet++ architecture trained pairs of 159 760 uncharged organic compounds from GEOM data set r2SCAN-3c reference level. Instead predicting only single molecules, this captures relative differences between conformers, leading to significant improvement overall conformational ranking, outperforming GFN-FF GFN2-xTB. Thereby, RMSD difference two can be reduced 5.65 0.71 kcal mol–1 test set, allowing correctly identify up 81% all lowest lying (GFN-FF: 10%, GFN2-xTB: 47%). The ConfRank cost-effective, scalable deployment both CPU GPU, achieving accelerations 2 orders magnitude compared Out-of-sample investigations CREST-generated ensembles QM9 taken an extended GMTKN55 show promising results robustness approach. correlation coefficient such as Spearman improved 0.90 0.39, 0.84) reducing probability incorrect sign flip in comparison 32 7%. On subsets MAD (RMSD) could almost 62% (58%) average 30% (29%). Moreover, exemplary case study vancomycin shows similar performance, indicating applicability larger (bio)molecular structures. Furthermore, motivate usage training theoretical perspective, highlighting while lead decline sample prediction absolute energies ML models, it significantly performance. models used available at https://github.com/grimme-lab/confrank.

Language: Английский

Citations

0