MetaScore: A Novel Machine-Learning-Based Approach to Improve Traditional Scoring Functions for Scoring Protein-Protein Docking Conformations

Yong Jung; Cunliang Geng; Alexandre M. J. J. Bonvin; Li C. Xue; Vasant G. Honavar

doi:10.3390/biom13010121

MetaScore: A Novel Machine-Learning-Based Approach to Improve Traditional Scoring Functions for Scoring Protein-Protein Docking Conformations

Yong Jung
, Cunliang Geng
, Alexandre M. J. J. Bonvin
, Li C. Xue^*
, Vasant G. Honavar

^*Corresponding author for this work

Research output: Contribution to conference › Paper › Academic

Abstract

Protein-protein interactions play a ubiquitous role in biological function. Knowledge of the three-dimensional (3D) structures of the complexes they form is essential for understanding the structural basis of those interactions and how they orchestrate key cellular processes. Computational docking has become an indispensable alternative to the expensive and time-consuming experimental approaches for determining the 3D structures of protein complexes. Despite recent progress, identifying near-native models from a large set of conformations sampled by docking-the so-called scoring problem-still has considerable room for improvement. We present MetaScore, a new machine-learning-based approach to improve the scoring of docked conformations. MetaScore utilizes a random forest (RF) classifier trained to distinguish near-native from non-native conformations using their protein-protein interfacial features. The features include physicochemical properties, energy terms, interaction-propensity-based features, geometric properties, interface topology features, evolutionary conservation, and also scores produced by traditional scoring functions (SFs). MetaScore scores docked conformations by simply averaging the score produced by the RF classifier with that produced by any traditional SF. We demonstrate that (i) MetaScore consistently outperforms each of the nine traditional SFs included in this work in terms of success rate and hit rate evaluated over conformations ranked among the top 10; (ii) an ensemble method, MetaScore-Ensemble, that combines 10 variants of MetaScore obtained by combining the RF score with each of the traditional SFs outperforms each of the MetaScore variants. We conclude that the performance of traditional SFs can be improved upon by using machine learning to judiciously leverage protein-protein interfacial features and by using ensemble methods to combine multiple scoring functions.

Original language	English
Number of pages	20
DOIs	https://doi.org/10.3390/biom13010121
Publication status	Published - Jan 2023

Bibliographical note

Publisher Copyright:
© 2023 by the authors.

Funding

Y.J. was supported in part by a research assistantship funded by a National Science Foundation grant 1640834 to V.H. The work of V.H. was supported in part by the National Center for Advancing Translational Sciences, National Institutes of Health through the grants UL1 TR000127 and TR002014, the National Science Foundation, through the grant 1640834, the Pennsylvania State University’s Institute for Computational and Data Sciences and the Center for Artificial Intelligence Foundations and Scientific Applications, the Edward Frymoyer Endowed Professorship in Information Sciences and Technology, the Dorothy Foehr Huck and J. Lloyd Huck Chair in Biomedical Data Sciences and Artificial Intelligence at Pennsylvania State University, and the Sudha Murty Distinguished Visiting Chair in Neurocomputing and Data Science funded by the Pratiksha Trust at the Indian Institute of Science. This work was also supported in part by the European H2020 e-Infrastructure grant BioExcel (grant no. 675728 and 823830) (A.M.J.J.B.). Financial support from the Netherlands Organisation for Scientific Research through an Accelerating Scientific Discovery (ASDI) from the Netherlands eScience Center (grant no. 027016G04) (L.C.X. and A.M.J.J.B.) and a Veni grant (grant no. 722.014.005) (L.C.X.) are acknowledged.

Funders	Funder number
Center for Artificial Intelligence Foundations and Scientific Applications
National Science Foundation	1640834
National Institutes of Health	UL1 TR000127, TR002014
National Center for Advancing Translational Sciences
Pennsylvania State University
Horizon 2020 Framework Programme	823830, 675728
Nederlandse Organisatie voor Wetenschappelijk Onderzoek	027016G04, 722.014.005

Keywords

machine learning
method combination
protein–protein docking
scoring functions

Access to Document

10.3390/biom13010121Licence: CC BY

biomolecules-13-00121-with-coverFinal published version, 881 KBLicence: CC BY

Cite this

@conference{5bf6c58bfb1940f292a60a252dc0ede9,

title = "MetaScore: A Novel Machine-Learning-Based Approach to Improve Traditional Scoring Functions for Scoring Protein-Protein Docking Conformations",

abstract = "Protein-protein interactions play a ubiquitous role in biological function. Knowledge of the three-dimensional (3D) structures of the complexes they form is essential for understanding the structural basis of those interactions and how they orchestrate key cellular processes. Computational docking has become an indispensable alternative to the expensive and time-consuming experimental approaches for determining the 3D structures of protein complexes. Despite recent progress, identifying near-native models from a large set of conformations sampled by docking-the so-called scoring problem-still has considerable room for improvement. We present MetaScore, a new machine-learning-based approach to improve the scoring of docked conformations. MetaScore utilizes a random forest (RF) classifier trained to distinguish near-native from non-native conformations using their protein-protein interfacial features. The features include physicochemical properties, energy terms, interaction-propensity-based features, geometric properties, interface topology features, evolutionary conservation, and also scores produced by traditional scoring functions (SFs). MetaScore scores docked conformations by simply averaging the score produced by the RF classifier with that produced by any traditional SF. We demonstrate that (i) MetaScore consistently outperforms each of the nine traditional SFs included in this work in terms of success rate and hit rate evaluated over conformations ranked among the top 10; (ii) an ensemble method, MetaScore-Ensemble, that combines 10 variants of MetaScore obtained by combining the RF score with each of the traditional SFs outperforms each of the MetaScore variants. We conclude that the performance of traditional SFs can be improved upon by using machine learning to judiciously leverage protein-protein interfacial features and by using ensemble methods to combine multiple scoring functions.",

keywords = "machine learning, method combination, protein–protein docking, scoring functions",

author = "Yong Jung and Cunliang Geng and Bonvin, \{Alexandre M. J. J.\} and Xue, \{Li C.\} and Honavar, \{Vasant G.\}",

note = "Publisher Copyright: {\textcopyright} 2023 by the authors.",

year = "2023",

month = jan,

doi = "10.3390/biom13010121",

language = "English",

}

TY - CONF

T1 - MetaScore: A Novel Machine-Learning-Based Approach to Improve Traditional Scoring Functions for Scoring Protein-Protein Docking Conformations

AU - Jung, Yong

AU - Geng, Cunliang

AU - Bonvin, Alexandre M. J. J.

AU - Xue, Li C.

AU - Honavar, Vasant G.

PY - 2023/1

Y1 - 2023/1

N2 - Protein-protein interactions play a ubiquitous role in biological function. Knowledge of the three-dimensional (3D) structures of the complexes they form is essential for understanding the structural basis of those interactions and how they orchestrate key cellular processes. Computational docking has become an indispensable alternative to the expensive and time-consuming experimental approaches for determining the 3D structures of protein complexes. Despite recent progress, identifying near-native models from a large set of conformations sampled by docking-the so-called scoring problem-still has considerable room for improvement. We present MetaScore, a new machine-learning-based approach to improve the scoring of docked conformations. MetaScore utilizes a random forest (RF) classifier trained to distinguish near-native from non-native conformations using their protein-protein interfacial features. The features include physicochemical properties, energy terms, interaction-propensity-based features, geometric properties, interface topology features, evolutionary conservation, and also scores produced by traditional scoring functions (SFs). MetaScore scores docked conformations by simply averaging the score produced by the RF classifier with that produced by any traditional SF. We demonstrate that (i) MetaScore consistently outperforms each of the nine traditional SFs included in this work in terms of success rate and hit rate evaluated over conformations ranked among the top 10; (ii) an ensemble method, MetaScore-Ensemble, that combines 10 variants of MetaScore obtained by combining the RF score with each of the traditional SFs outperforms each of the MetaScore variants. We conclude that the performance of traditional SFs can be improved upon by using machine learning to judiciously leverage protein-protein interfacial features and by using ensemble methods to combine multiple scoring functions.

AB - Protein-protein interactions play a ubiquitous role in biological function. Knowledge of the three-dimensional (3D) structures of the complexes they form is essential for understanding the structural basis of those interactions and how they orchestrate key cellular processes. Computational docking has become an indispensable alternative to the expensive and time-consuming experimental approaches for determining the 3D structures of protein complexes. Despite recent progress, identifying near-native models from a large set of conformations sampled by docking-the so-called scoring problem-still has considerable room for improvement. We present MetaScore, a new machine-learning-based approach to improve the scoring of docked conformations. MetaScore utilizes a random forest (RF) classifier trained to distinguish near-native from non-native conformations using their protein-protein interfacial features. The features include physicochemical properties, energy terms, interaction-propensity-based features, geometric properties, interface topology features, evolutionary conservation, and also scores produced by traditional scoring functions (SFs). MetaScore scores docked conformations by simply averaging the score produced by the RF classifier with that produced by any traditional SF. We demonstrate that (i) MetaScore consistently outperforms each of the nine traditional SFs included in this work in terms of success rate and hit rate evaluated over conformations ranked among the top 10; (ii) an ensemble method, MetaScore-Ensemble, that combines 10 variants of MetaScore obtained by combining the RF score with each of the traditional SFs outperforms each of the MetaScore variants. We conclude that the performance of traditional SFs can be improved upon by using machine learning to judiciously leverage protein-protein interfacial features and by using ensemble methods to combine multiple scoring functions.

KW - machine learning

KW - method combination

KW - protein–protein docking

KW - scoring functions

UR - https://www.scopus.com/pages/publications/85146752195

U2 - 10.3390/biom13010121

DO - 10.3390/biom13010121

M3 - Paper

C2 - 36671507

ER -

MetaScore: A Novel Machine-Learning-Based Approach to Improve Traditional Scoring Functions for Scoring Protein-Protein Docking Conformations

Abstract

Bibliographical note

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this