Discrimination and Prediction of Protein-Protein Binding Affinity Using Deep Learning Approach

Systematic Investigation of Machine Learning on Limited Data: A Study on Predicting Protein-Protein Binding Strength

“…We performed a comprehensive comparison between our models and eight other state-of-the-art methods used to calculate the binding energy, including PRODIGY 44,45 , PPI-Affinity 38 , PPA_Pred2 16 , Minpredictor 42 , ISLAND 43 , FoldX 75 , Rosetta 76 , and MMPBSA 8,77 . PPA_Pred2 and ISLAND are sequence-based approaches, while the rest are structure-based methods.…”

“…However, this approach has faced a bottleneck, leading to a lack of substantial improvement in predictive performance in recent years 14 . Despite the introduction of deep learning approaches in this field, their performance improvement has been severely hampered [15][16][17] , primarily due to the limited availability of training data. Continued efforts are required to overcome the challenges posed by limited data in machine learning.…”

“…One such method is statistical potentials, which uses the observed relative positions of atoms or residues in experimental structures to infer a potential of mean force 36,37 . Another approach that has gained increasing popularity over the past decade is machine learning, where energy functions are determined through regression against experimentally measured binding affinities 16,[38][39][40][41][42][43][44][45][46] . However, the prediction accuracy of currently available methods remains limited.…”

Systematic investigation of machine learning on limited data: A study on predicting protein-protein binding strength

“…However, this approach faced a bottleneck, leading to a lack of substantial improvement in predictive performance in recent years [14] . Despite the introduction of deep learning approaches in this field, their performance improvement has been severely hampered [15] , [16] , [17] , primarily due to the limited availability of training data. Continued efforts are required to overcome the challenges posed by limited data in machine learning.…”

“…One such method is statistical potentials, which use the observed relative positions of atoms or residues in experimental structures to infer a potential of mean force [36] , [37] . Another approach that has gained increasing popularity over the past decade is machine learning, where energy functions are determined through regression against experimentally measured binding affinities [16] , [38] , [39] , [40] , [41] , [42] , [43] , [44] , [45] , [46] . However, the prediction accuracy of currently available methods remains limited.…”