ISSEC

ISSEC infers contacts among protein secondary structure elements (SSEs) from predicted inter-residue contact maps using deep object-detection and deep convolutional neural network approaches to support tertiary-structure topology analysis and protein folding studies.

Key Features:

• Deep learning approach: Applies deep object-detection concepts and deep convolutional neural networks to extract high-level features from predicted inter-residue contact maps.
• Pattern recognition: Identifies rectangular pattern regions in inter-residue contact maps that correspond to contacting SSEs.
• Non-reliance on predefined SSEs: Dynamically enumerates multiple candidate rectangular regions without requiring predefined secondary structure element boundaries.
• Candidate enumeration and scoring: Enumerates multiple candidate rectangular regions and assigns a confidence score to each based on pattern characteristics.
• Confidence-based selection: Uses a greedy strategy to select non-overlapping regions with high confidence scores for final inter-SSE contact inference.

Scientific Applications:

• Inter-SSE contact prediction: Improves inference of contacts among SSEs and has demonstrated superior performance relative to existing approaches.
• Robust contact inference: Infers SSE contacts without predefined SSE boundaries, mitigating errors from secondary structure prediction and noise in predicted inter-residue contacts.
• Tertiary structure modeling: Enhances accuracy of inter-residue contact predictions and downstream tertiary structure modeling.
• Protein folding and topology analysis: Supports studies of protein folding mechanisms and analysis of protein tertiary-structure topology.

Methodology:

Analyze predicted inter-residue contact maps for rectangular regions, extract high-level features using deep convolutional techniques, enumerate multiple candidate rectangular regions, assign confidence scores based on pattern characteristics, and select non-overlapping high-confidence regions using a greedy algorithm.