DeepIDP-2L

DeepIDP-2L predicts intrinsically disordered regions (IDRs) in proteins, distinguishing long disordered regions (LDRs) and short disordered regions (SDRs) to improve characterization of disorder-related structural and functional properties.

Key Features:

• Two-layer predictive architecture: Separately extracts and integrates features for LDRs and SDRs using a hierarchical two-layer model.
• Hierarchical attention network: Captures distribution pattern features specific to long disordered regions (LDRs).
• Convolutional attention network: Identifies local correlation features specific to short disordered regions (SDRs).
• Feature transformation and fusion: Transforms extracted features into a new feature space using machine learning techniques for downstream prediction.
• Convolutional network and Bi-LSTM integration: Employs a convolutional network and bidirectional long short-term memory (Bi-LSTM) to capture local and long-range information.
• Robust performance across datasets: Demonstrates stable performance across independent test sets with varying ratios of SDRs and LDRs.

Scientific Applications:

• IDR annotation: Annotation and identification of intrinsically disordered regions in protein sequences.
• Disorder-type differentiation: Differentiating long and short disordered regions to inform structure–function analyses.
• Predictor evaluation: Comparative evaluation and benchmarking of IDR prediction methods across datasets with varying SDR/LDR composition.

Methodology:

First layer: a hierarchical attention network captures LDR distribution pattern features and a convolutional attention network captures SDR local correlation features; second layer: extracted features are transformed into a new feature space and processed by a convolutional network and bidirectional long short-term memory (Bi-LSTM) to capture local and long-range information.