Abstract
Accurate high-throughput prediction of protein-protein interactions (PPIs) is essential for mapping cellular mechanisms and prioritizing experimental validation. Current graph-based methods often rely on discrete message passing, suffer from representation over-smoothing, and provide poorly calibrated confidence scores. We present LNGCN, a distance-aware continuous-time graph protocol that integrates residue-level structural graphs with liquid neural dynamics to model spatially heterogeneous interaction patterns. Residue radial distance is used as an explicit driving signal for continuous graph evolution, while hierarchical calibration converts raw model outputs into interpretable interaction probabilities. Across balanced, highly imbalanced, and cross-species benchmarks, LNGCN achieved robust predictive performance. Importantly, the calibrated scores supported biologically coherent prioritization in the FGF23-FGFR1c- α -Klotho complex, SHP2-associated signaling interactions and Tdk1 oligomeric-state-dependent binding. In a TPR-centered experimental case study, LNGCN recovered known TPR-associated partners and prioritized ELAVL1-TPR and RALY-TPR, whose physical interactions were subsequently confirmed via experimental validation. These results indicate that LNGCN can serve as a practical prioritization protocol for PPI candidates.
Full text
1,367 characters
· extracted from
oa-doi-fallback
· click to expand
Abstract
High-throughput accurate protein-protein interaction (PPI) prediction is foundational to systems-level biological understanding, disease mechanism dissection, and structure-based drug discovery. Traditional graph convolutional networks (GCNs) are limited by discrete information propagation, layer-wise representation homogenization, and absent continuous-time state evolution, failing to capture residues’ 3D spatial hierarchical dynamic binding patterns. We present LNGCN, a hybrid framework integrating liquid neural networks with GCNs, which encodes residue radial distances as node-level driving terms for continuous updates with hierarchical probabilistic calibration. On standard benchmarks, LNGCN achieves 90% relative AUPRC improvement over PIPR, outperforms RF2-PPI on 1 : 10 imbalanced datasets, and retains 0.9324 AUPRC on held-out yeast test data. LNGCN further demonstrates biological utility in phosphorylation-dependent SHP2 signaling, FGF23-FGFR1c-α-Klotho ternary assembly, Tdk1 oligomeric-state-dependent interactions, and experimentally validated TPR-mediated candidates. By capturing state-dependent interaction changes, LNGCN provides a scalable framework for PPI screening, candidate prioritization, and future residue-level dynamic PPI trajectory modeling.
Competing Interest Statement
The authors have declared no competing interest.
Text is read by the "Ask this paper" AI Q&A widget below.
Extraction quality varies by source — PMC NXML preserves structure
cleanly, OA-HTML may include some navigation residue, and OA-PDF can
have broken hyphenation. The publisher copy
(via DOI)
is the canonical version.