TY - JOUR

T1 - PCANN Program for Structure-Based Prediction of Protein–Protein Binding Affinity: Comparison With Other Neural-Network Predictors

AU - Лебеденко, Ольга Олеговна

AU - Половинкин, Михаил Сергеевич

AU - Казовская, Анастасия Александровна

AU - Скрынников, Николай Русланович

N1 - Lebedenko O.O., Polovinkin M.S., Kazovskaia A.A., Skrynnikov N.R. PCANN Program for Structure-Based Prediction of Protein–Protein Binding Affinity: Comparison With Other Neural-Network Predictors // Proteins: Structure, Function, and Bioinformatics – 2025. P.1-9.

PY - 2025/9/1

Y1 - 2025/9/1

N2 - In this communication, we introduce a new structure-based affinity predictor for protein–protein complexes. This predictor, dubbed PCANN (Protein Complex Affinity by Neural Network), uses the ESM-2 language model to encode the information about protein binding interfaces and graph attention network (GAT) to parlay this information into (Formula presented.) predictions. In the tests employing two previously unused literature-extracted datasets, PCANN performed better than the best of the publicly available predictors, BindPPI, with mean absolute error (MAE) of 1.3 versus 1.4 kcal/mol. Further progress in the development of (Formula presented.) predictors using deep learning models is faced with two problems: (i) the amount of experimental data available to train and test new predictors is limited and (ii) the available (Formula presented.) data are often not very accurate and lack internal consistency with respect to measurement conditions. These issues can be potentially addressed through an AI-leveraged literature search followed by careful human curation and by introducing additional parameters to account for variations in experimental conditions.

AB - In this communication, we introduce a new structure-based affinity predictor for protein–protein complexes. This predictor, dubbed PCANN (Protein Complex Affinity by Neural Network), uses the ESM-2 language model to encode the information about protein binding interfaces and graph attention network (GAT) to parlay this information into (Formula presented.) predictions. In the tests employing two previously unused literature-extracted datasets, PCANN performed better than the best of the publicly available predictors, BindPPI, with mean absolute error (MAE) of 1.3 versus 1.4 kcal/mol. Further progress in the development of (Formula presented.) predictors using deep learning models is faced with two problems: (i) the amount of experimental data available to train and test new predictors is limited and (ii) the available (Formula presented.) data are often not very accurate and lack internal consistency with respect to measurement conditions. These issues can be potentially addressed through an AI-leveraged literature search followed by careful human curation and by introducing additional parameters to account for variations in experimental conditions.

KW - protein binding affinity

KW - graph attention network

KW - language model

KW - ESM-2 language model

KW - Kd prediction program

KW - comparison of Kd predictors

KW - deep learning

KW - graph attention network

KW - protein binding databases

KW - protein–protein binding

KW - Neural Networks, Computer

KW - Humans

KW - Computational Biology/methods

KW - Deep Learning

KW - Proteins/chemistry

KW - Algorithms

KW - Databases, Protein

KW - Protein Binding

KW - Protein Conformation

KW - Software

KW - Binding Sites

UR - https://www.mendeley.com/catalogue/f000f359-cd4f-3022-9666-1f935f07366f/

UR - https://elibrary.ru/item.asp?id=81941876

U2 - 10.1002/prot.26821

DO - 10.1002/prot.26821

M3 - Article

C2 - 40116085

VL - 93

SP - 1498

EP - 1506

JO - Proteins: Structure, Function and Bioinformatics

JF - Proteins: Structure, Function and Bioinformatics

SN - 0887-3585

IS - 9

ER -