Related Experiment Video
Updated: Jan 18, 2026
06:50
Author Spotlight: A Computational Approach to Decipher Amino Acid Preferences in Multispecific Protein-Protein Interactions
Published on: January 26, 2024
2.5K
Protein Language Models and Structure-Based Machine Learning for Prediction of Allosteric Binding Sites in Protein
Biorxiv : the Preprint Server for Biology
|January 16, 2026
Summary
Predicting allosteric binding sites in protein kinases using AI is challenging due to their cryptic nature. Protein frustration analysis reveals that allosteric sites, unlike orthosteric sites, have neutral mutational constraints, explaining AI performance disparities.
Area of Science:
- Biochemistry and Structural Biology
- Computational Biology and Cheminformatics
- Drug Discovery and Medicinal Chemistry
Background:
- Identifying allosteric binding sites is crucial for drug discovery but remains a significant challenge, especially in protein kinases.
- Allosteric sites are often structurally cryptic, evolutionarily non-conserved, and sparsely populated, making them difficult to predict.
- Existing computational methods struggle with the accurate prediction of these challenging sites.
Purpose of the Study:
- To systematically analyze the performance of a fine-tuned protein language model (PLM) and a structure-based approach (P2Rank) for predicting orthosteric and allosteric binding sites in human kinases.
- To investigate the mechanistic basis for performance discrepancies between orthosteric and allosteric site prediction using local frustration analysis.
- To reframe AI performance in binding site prediction as a reflection of functional design and protein frustration.
Main Methods:
- A curated dataset of 453 human kinase-ligand complexes spanning five inhibitor classes was used.
- A pretrained protein language model (ESM2-650M) was fine-tuned for binding site prediction.
- Sequence-based PLM and structure-based P2Rank were employed for site identification.
- Large-scale local frustration analysis was integrated to interpret prediction discrepancies.
Main Results:
- Both PLM and P2Rank achieved high performance on orthosteric sites (AUPR = 0.64-0.76).
- PLM performance significantly dropped for allosteric sites (AUPR = 0.06), despite moderate ranking ability (AUROC = 0.70).
- Local frustration analysis revealed orthosteric sites are enriched in minimally frustrated residues, while allosteric sites exhibit neutral mutational frustration, indicating evolutionary permissiveness.
Conclusions:
- The performance gap of AI models in predicting allosteric versus orthosteric kinase sites is linked to their distinct underlying biophysical properties, specifically mutational constraints.
- Protein frustration serves as an explainable AI framework to rationalize AI performance in binding site prediction.
- Understanding these differences is key to improving structure-based drug discovery for allosteric modulators.
Related Concept Videos
02:04Protein-protein Interfaces
14.4K
Many proteins form complexes to carry out their functions, making protein-protein interactions (PPIs) essential for an organism's survival. Most PPIs are stabilized by numerous weak noncovalent chemical forces. The physical shape of the interfaces determines the way two proteins interact. Many globular proteins have closely-matching shapes on their surfaces, which form a large number of weak bonds. Additionally, many PPIs occur between two helices or between a surface cleft and a...
14.4K
02:04Protein-Protein Interfaces
4.4K
4.4K
02:40Ligand Binding Sites
14.9K
Proteins are dynamic macromolecules that carry out a wide variety of essential processes; however, the activities of most proteins depend on their interactions with other molecules or ions, known as ligands.
Protein-ligand interactions are quite specific; even though numerous potential ligands surround a cellular protein at any given time, only a particular ligand can bind to that protein. Moreover, a ligand binds only to a dedicated area on the surface of the protein, known as the...
Protein-ligand interactions are quite specific; even though numerous potential ligands surround a cellular protein at any given time, only a particular ligand can bind to that protein. Moreover, a ligand binds only to a dedicated area on the surface of the protein, known as the...
14.9K
02:40Ligand Binding Sites
8.6K
8.6K
01:49Conserved Binding Sites
5.0K
Many proteins’ biological role depends on their interactions with their ligands, small molecules that bind to specific locations on the protein known as ligand-binding sites. Ligand-binding sites are often conserved among homologous proteins as these sites are critical for protein function.
Binding sites are often located in large pockets, and if their location on a protein’s surface is unknown, it can be predicted using various approaches. The energetic method computationally...
Binding sites are often located in large pockets, and if their location on a protein’s surface is unknown, it can be predicted using various approaches. The energetic method computationally...
5.0K
01:19Allosteric Proteins-ATCase
6.5K
Binding sites linkages can regulate a protein's function. For example, enzyme activity is often regulated through a feedback mechanism where the end product of the biochemical process serves as an inhibitor.
Aspartate transcarbamoylase (ATCase) is a cytosolic enzyme that catalyzes the condensation of L-aspartate and carbamoyl phosphate to N-carbamoyl-L-aspartate. This reaction is the first step in pyrimidine biosynthesis. UTP and CTP, the end products of the pyrimidine synthesis...
Aspartate transcarbamoylase (ATCase) is a cytosolic enzyme that catalyzes the condensation of L-aspartate and carbamoyl phosphate to N-carbamoyl-L-aspartate. This reaction is the first step in pyrimidine biosynthesis. UTP and CTP, the end products of the pyrimidine synthesis...
6.5K