Greta Grassmann, Lorenzo Di Rienzo, Giancarlo Ruocco, Mattia Miotto, Edoardo Milanetti
{"title":"分子表面互补性的紧凑评估增强了神经网络辅助预测关键结合残基的能力","authors":"Greta Grassmann, Lorenzo Di Rienzo, Giancarlo Ruocco, Mattia Miotto, Edoardo Milanetti","doi":"arxiv-2407.20992","DOIUrl":null,"url":null,"abstract":"Predicting interactions between biomolecules, such as protein-protein\ncomplexes, remains a challenging problem. Despite the many advancements done so\nfar, the performances of docking protocols are deeply dependent on their\ncapability of identify binding regions. In this context, we present a novel\napproach that builds upon our previous works modeling protein surface patches\nvia sets of orthogonal polynomials to identify regions of high\nshape/electrostatic complementarity. By incorporating another key binding\nproperty, such as the balance between hydrophilic and hydrophobic\ncontributions, we define new binding matrices that serve an effective inputs\nfor training a neural network. Our approach also allows for the quantitative\ndefinition of a typical binding site area - approximately 10\\AA~in radius -\nwhere hydrophobic contribution and shape complementarity, which reflects the\nLennard-Jones interaction, are maximized. Using this new architecture, CIRNet\n(Core Interacting Residues Network), we achieve an accuracy of approximately\n0.82 in identifying pairs of core interacting residues on a balanced dataset.\nIn a blind search for core interacting residues, CIRNet distinguishes these\nfrom decoys with a ROC AUC of 0.72. This protocol can enahnce docking\nalgorithms by rescaling the proposed poses. When applied to the top ten models\nfrom three popular docking server, CIRNet improves docking outcomes, reducing\nthe the average RMSD between the refined poses and the native state by up to\n58%.","PeriodicalId":501022,"journal":{"name":"arXiv - QuanBio - Biomolecules","volume":"74 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Compact assessment of molecular surface complementarities enhances neural network-aided prediction of key binding residues\",\"authors\":\"Greta Grassmann, Lorenzo Di Rienzo, Giancarlo Ruocco, Mattia Miotto, Edoardo Milanetti\",\"doi\":\"arxiv-2407.20992\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Predicting interactions between biomolecules, such as protein-protein\\ncomplexes, remains a challenging problem. Despite the many advancements done so\\nfar, the performances of docking protocols are deeply dependent on their\\ncapability of identify binding regions. In this context, we present a novel\\napproach that builds upon our previous works modeling protein surface patches\\nvia sets of orthogonal polynomials to identify regions of high\\nshape/electrostatic complementarity. By incorporating another key binding\\nproperty, such as the balance between hydrophilic and hydrophobic\\ncontributions, we define new binding matrices that serve an effective inputs\\nfor training a neural network. Our approach also allows for the quantitative\\ndefinition of a typical binding site area - approximately 10\\\\AA~in radius -\\nwhere hydrophobic contribution and shape complementarity, which reflects the\\nLennard-Jones interaction, are maximized. Using this new architecture, CIRNet\\n(Core Interacting Residues Network), we achieve an accuracy of approximately\\n0.82 in identifying pairs of core interacting residues on a balanced dataset.\\nIn a blind search for core interacting residues, CIRNet distinguishes these\\nfrom decoys with a ROC AUC of 0.72. This protocol can enahnce docking\\nalgorithms by rescaling the proposed poses. When applied to the top ten models\\nfrom three popular docking server, CIRNet improves docking outcomes, reducing\\nthe the average RMSD between the refined poses and the native state by up to\\n58%.\",\"PeriodicalId\":501022,\"journal\":{\"name\":\"arXiv - QuanBio - Biomolecules\",\"volume\":\"74 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-07-30\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - QuanBio - Biomolecules\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2407.20992\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - QuanBio - Biomolecules","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2407.20992","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Compact assessment of molecular surface complementarities enhances neural network-aided prediction of key binding residues
Predicting interactions between biomolecules, such as protein-protein
complexes, remains a challenging problem. Despite the many advancements done so
far, the performances of docking protocols are deeply dependent on their
capability of identify binding regions. In this context, we present a novel
approach that builds upon our previous works modeling protein surface patches
via sets of orthogonal polynomials to identify regions of high
shape/electrostatic complementarity. By incorporating another key binding
property, such as the balance between hydrophilic and hydrophobic
contributions, we define new binding matrices that serve an effective inputs
for training a neural network. Our approach also allows for the quantitative
definition of a typical binding site area - approximately 10\AA~in radius -
where hydrophobic contribution and shape complementarity, which reflects the
Lennard-Jones interaction, are maximized. Using this new architecture, CIRNet
(Core Interacting Residues Network), we achieve an accuracy of approximately
0.82 in identifying pairs of core interacting residues on a balanced dataset.
In a blind search for core interacting residues, CIRNet distinguishes these
from decoys with a ROC AUC of 0.72. This protocol can enahnce docking
algorithms by rescaling the proposed poses. When applied to the top ten models
from three popular docking server, CIRNet improves docking outcomes, reducing
the the average RMSD between the refined poses and the native state by up to
58%.