Riedlova, K. | bioRxiv Science

bioRxiv2026

Decoding Allosteric Grammar with Explainable AI Integrating Protein Language Models and Energy Landscape Analysis: Neutral Frustration at Allosteric Binding Sites Encodes Regulatory Versatility in Protein Kinases

Allosteric regulation enables protein kinases to integrate diverse cellular signals, yet the energetic organization principles that encode spatial and evolutionary diversity of regulatory binding sites lack a complete understanding. We introduce an explainable artificial intelligence (AI) framework that uses protein language models (PLMs) not as predictive endpoints, but as diagnostic probes of biophysical organization of regulatory regions. By integrating PLM-based binding site predictions with the energy landscape-based frustration analysis, we determine that the detectability of orthosteric and allosteric binding sites reflects their energetic embedding within the protein energy landscape. Using PLM predictions as unbiased probes across a structurally and functionally diverse dataset of 453 human kinases, we observe a striking and reproducible dichotomy in predictive behavior: orthosteric ATP-binding pockets are detected with high confidence, whereas allosteric sites consistently evade robust detection. Orthosteric catalytic sites reside within minimally frustrated, optimized energetic regions that are consistently detected with high confidence. In contrast, allosteric sites are enriched in neutrally frustrated zones, producing diffuse and context-dependent predictions, revealing that the "allosteric blind spot" arises from intrinsic biophysical design rather than algorithmic limitations. Atomic-resolution analysis of ABL kinase spanning multiple conformational states and complexes bound to diverse ligands provides mechanistic validation of this principle. The myristoyl allosteric pocket in ABL remains neutrally frustrated across complexes with physiological ligands, chemically diverse modulators, from allosteric inhibitors to activators, and conformations engaged with SH2-SH3 regulatory domains. We propose that allosteric sites are encoded in persistent neutrally frustrated regions optimized for context-dependent regulatory modulation. By using explainable AI to interrogate the energetic architecture of protein kinases, this work reveals how the organization of the protein energy landscape shapes functional plasticity and algorithmic detectability of regulatory binding sites.

biophysics↗

bioRxiv2026

Protein Language Models and Structure-Based Machine Learning for Prediction of Allosteric Binding Sites in Protein Kinases: An Explainable AI Framework Grounded in Energy Landscape-Encoded Frustration

Reliable identification of allosteric binding sites remains a major bottleneck in structure-based drug discovery, particularly in protein kinase families where such sites are often structurally cryptic, evolutionarily non-conserved, and sparsely populated. In this work, we present a systematic analysis of binding site prediction across a rigorously curated dataset of human kinase-ligand complexes, encompassing 453 kinases and spanning five inhibitor classes: Type I, Type I.5, and Type II (orthosteric ATP-competitive) and Type III/IV (non-ATP allosteric) modulators. We employed the pretrained protein language model (PLM) ESM2-650M model that was fine-tuned for prediction of protein-ligand binding sites by replacing the original masked language modeling head with a token-level classification head that acts as a projection layer that maps the high-dimensional latent representation of each residue to a scalar probability score for a given protein residue to be part of the binding site. We employed this fine-tuned sequence-based PLM and structure-based detection approach P2Rank for identification of orthosteric and allosteric binding sites in protein kinases. Our analysis reveals a stark performance divergence: while both methods achieve high precision-recall (AUPR = 0.64-0.76) on orthosteric sites, PLM performance collapses on allosteric sites (AUPR = 0.06), despite retaining moderate ranking ability (AUROC = 0.70). This deficit persists even after strict control for sequence similarity, structural redundancy, and extreme class imbalance (allosteric residues constitute <3% of the kinase domain). To mechanistically interpret this discrepancy, we integrate large-scale local frustration analysis, a physics-based framework derived from energy landscape theory that quantifies the energetic stability of residue-residue interactions under mutational and conformational perturbations. We find that, although the global frustration landscape is conserved across kinase states dominated by neutral frustration (55-75% of residues), local binding sites exhibit fundamentally distinct mutational constraints. Orthosteric pockets are enriched in minimally frustrated residues, whereas allosteric sites are characterized by neutral mutational frustration, indicating evolutionary permissiveness and sequence degeneracy. This study reframes the performance of AI approaches in predicting protein binding sites as a reflection of functional design that can be rationalized through lens of the landscape-encoded protein frustration as an explainable AI framework.

biophysics↗