Decoding disorder: Advances in predicting intrinsically disordered proteins

Intrinsically disordered proteins (IDPs) and intrinsically disordered regions (IDRs) play critical roles in cellular signaling, transcriptional regulation, and dynamic protein-protein interactions. Owing to their lack of stable three-dimensional structure under physiological conditions, IDPs remain difficult to characterize experimentally. This limitation has driven the development of various computational methods for the high-throughput prediction IDPs. This mini-review focuses on recent advances reported in 2025, spanning four major developments: (i) ensemble deep-learning frameworks like IDP-EDL that integrate task-specific predictors; (ii) transformer-based language models, including ProtT5 and ESM-2, which offer rich residue-level embeddings for disorder and MoRF prediction; (iii) multi-feature fusion models such as FusionEncoder that combine evolutionary, physicochemical, and semantic features to improve boundary accuracy; and (iv) hybrid approaches that integrate AlphaFold-predicted distance restraints with molecular dynamics to generate structural ensembles. We also review benchmarking efforts from the CAID2 initiative and discuss emerging applications of IDP prediction in proteomics, including post-translational modification mapping, interactome analysis, and therapeutic targeting. Finally, we outline future directions involving experimental integration, functional annotation, and explainable AI. These advancements represent a significant leap in the computational understanding of protein disorder.