metapredict

metapredict predicts per-residue intrinsic disorder in protein sequences using a bidirectional recurrent neural network trained on consensus disorder scores.

Key Features:

• Deep Learning-Based Prediction: Employs a bidirectional recurrent neural network (RNN) to predict per-residue consensus disorder scores.
• Consensus Score Integration: Integrates outputs from multiple independent disorder predictors to produce per-residue consensus values reflecting the number of tools calling a residue disordered.
• Efficiency and Speed: Performs proteome-scale disorder prediction within minutes.

Scientific Applications:

• Experimental Design and Interpretation: Guides experimental design, interpretation of results, and hypothesis development regarding protein function.
• Single-Protein and Proteome-Level Analysis: Enables characterization of intrinsic disorder at both single-protein and proteome scales to support studies of protein function and biological processes.

Methodology:

Trained on consensus disorder scores derived from 12 proteomes, metapredict employs a bidirectional RNN architecture, integrates outputs from multiple independent disorder predictors to produce per-residue consensus scores, and was benchmarked using two orthogonal approaches.