DisoLipPred

DisoLipPred predicts disordered lipid-binding residues (DLBRs) from protein sequences to identify intrinsically disordered regions that mediate protein–lipid interactions.

Key Features:

• Deep Bidirectional Recurrent Network: Employs a deep bidirectional recurrent neural network to capture sequential dependencies in protein sequences.
• Transfer Learning: Applies transfer learning using pre-trained models on related tasks to improve predictive accuracy with limited DLBR-specific training data.
• Bypass Module: Includes a bypass module that excludes residues likely to be structured to focus predictions on disordered regions.
• Expanded Inputs: Integrates expanded input features including physicochemical properties relevant to protein–lipid interactions.
• Performance and Complementarity: Validated on independent datasets and the yeast proteome, showing high accuracy, outperforming indirect DLBR-identifying tools, and complementing transmembrane-region predictors.

Scientific Applications:

• IDP annotation: Annotating disordered lipid-binding residues in intrinsically disordered proteins (IDPs) to study their roles in cellular functions.
• Mechanistic studies: Aiding elucidation of protein–lipid interaction mechanisms implicated in biological processes and pathologies.
• Complementary structural analysis: Complementing transmembrane-region predictors to provide a more holistic view of protein structure–function relationships.

Methodology:

Uses a deep bidirectional recurrent neural network with transfer learning, a bypass module to exclude structured residues, and expanded physicochemical input features.