AI-based prediction of intrinsically disordered protein regions

This video summarizes the paper “flDPnn: Accurate Intrinsic Disorder Prediction with Putative Propensities of Disorder Functions” by Gang Hu, et al. (Nature Communications, 2021). The authors developed flDPnn, a deep learning framework for predicting intrinsically disordered protein regions (IDRs) and their potential functions directly from sequence data. Using innovations such as protein-level feature encoding and integrated disorder-function profiles, flDPnn achieved state-of-the-art accuracy in the Critical Assessment of Intrinsic Disorder Prediction (CAID) benchmark, outperforming previous models in both speed and precision. Unlike traditional predictors that focus solely on disorder, flDPnn also assigns propensities for four key disorder-associated functions (protein-, DNA-, and RNA-binding, and linker regions), providing a unified view of sequence disorder and function. This tool represents a major step forward in mapping the functional “dark proteome” and understanding how structural flexibility contributes to cellular regulation.