AI for plant protein-protein interactions prediction

Protein-Protein interactions (PPIs) wire plant cells, assembling metabolons, routing signals, and coordinating organelle crosstalk. We review experimental platforms and the computational signals long used to predict PPIs. While experimental platforms and traditional computational approaches have long been employed for PPIs prediction, recent advances in artificial intelligence offer unprecedented opportunities to map plant interactomes comprehensively. To provide a systematic overview, we categorize current methodologies into four thematic families: (i) sequence-centric predictors utilizing protein language models to extract evolutionary features; (ii) structure-based predictors integrating coevolutionary signals to reconstruct 3D complex arrangements; (iii) network-level learners employing graph architectures to capture global interactome topology; and (iv) geometric and generative methods leveraging symmetry-aware networks for specific site identification and de novo design. Despite rapid gains, plant applications are constrained by paralog expansion, compartmentalization, dynamic microenvironments, and the sparse availability of gold standards in the field. Next-generation plant AI PPI models should be organelle-aware, multimodal, rigorously benchmarked, structure-gated, and condition-validated.