DisoMine

DisoMine predicts intrinsically disordered protein regions (IDRs) from single protein sequences by leveraging recurrent neural networks and sequence-derived biophysical property predictions for structural and functional characterization of proteins.

Key Features:

• Recurrent Neural Networks: Employs recurrent neural networks (RNNs) to model sequence information for disorder prediction.
• Biophysical Property Integration: Uses predictions of protein dynamics, secondary structure, and early folding as inputs to the predictor.
• Emergent Sequence-dependent Properties: Focuses on sequence-dependent emergent properties such as protein backbone dynamics rather than solely on raw amino acid sequence.
• Single-sequence Input: Requires only a single protein sequence, enabling analysis of orphan or poorly characterized proteins.
• Comparative Performance: Demonstrates performance competitive with disorder predictors that rely on evolutionary information.
• Large-scale Applicability: Optimized for speed and efficiency to support large-scale screenings.

Scientific Applications:

• Structural Biology: Identification and characterization of IDRs to inform studies of protein structure and dynamics.
• Functional Annotation: Functional inference for orphan or poorly characterized proteins through disorder prediction.
• Proteome-scale Screening: Large-scale identification of disordered regions across many proteins for comparative analyses.
• Study of Cellular Processes: Investigation of IDR roles in protein interactions and cellular processes.

Methodology:

DisoMine applies recurrent neural networks to sequence-derived predictions of biophysical properties—including protein dynamics, secondary structure, and early folding—emphasizing sequence-dependent backbone dynamics rather than evolutionary information.