DFLpred

DFLpred predicts disordered flexible linkers (DFLs) in protein sequences by assigning per-residue propensity scores to identify intrinsically disordered, extended linker regions that act as flexible spacers between domains or structured regions.

Key Features:

• Propensity Scoring: Assigns a numeric score to each residue indicating its propensity to form a disordered flexible linker.
• Empirical Feature Set: Uses a small set of empirically selected features quantifying propensities related to secondary structure, disordered regions, and structured regions processed through a fast linear model.
• High-Throughput Capability: Can process entire proteomes in under one hour on a single CPU.
• Performance Metrics: Achieved an area under the ROC curve (AUC) of 0.715 on an independent test dataset of proteins with low sequence identity, outperforming methods that predict flexible linkers, flexible residues, intrinsically disordered residues, or their combinations.
• Proteome-Wide Insights: Applied to the human proteome, estimated that ~10% of proteins contain >30% DFL residues and identified ~6,000 DFL regions of at least 30 consecutive residues.

Scientific Applications:

• Multi-domain protein analysis: Identification of DFLs aids studies of domain organization and linker-mediated function in multi-domain proteins.
• Protein dynamics and interactions: Enables analysis of protein dynamics and interaction mechanisms mediated by intrinsically disordered linkers.
• Proteome-scale surveys: Supports proteome-wide analyses of disorder and linker prevalence and comparative studies across proteins.

Methodology:

Uses empirically selected sequence-derived features related to secondary structure and disorder that are input to a fast linear model.