SPOT-Disorder

SPOT-Disorder predicts intrinsically disordered regions in protein sequences using deep bidirectional LSTM recurrent neural networks to identify disorder and functional sites for structural genomics, molecular biology, and drug discovery applications.

Key Features:

• Advanced Neural Network Architecture: Uses deep bidirectional LSTM recurrent neural networks to capture long-range, non-local interactions between residues that are structural but not sequence neighbors.
• No Separate Short/Long Region Training: Predicts intrinsic disorder without requiring separate training on short and long disordered regions.
• Improved Performance versus Window-Based Methods: Captures non-local interactions enabling performance superior to window-based neural network methods such as SPINE-D.
• Benchmarking: Validated on datasets including CASP and over 10,000 annotated proteins from MobiDB, demonstrating improved accuracy in predicting intrinsic disorder and functional sites.
• Functional Site Identification: Detects functional sites within disordered regions of proteins.

Scientific Applications:

• Drug Discovery: Supports identification of disordered regions and functional sites that can be considered in therapeutic target design.
• Structural Genomics: Guides experimental structure determination by predicting regions of intrinsic disorder that affect crystallization and structure modeling.
• Molecular Biology and Functional Annotation: Aids studies of protein function by mapping intrinsically disordered proteins (IDPs) and their functional sites.
• Target Identification: Assists in pinpointing potential targets for therapeutic intervention within disordered regions.

Methodology:

Implementation of deep bidirectional LSTM recurrent neural networks to predict intrinsic disorder by capturing long-range, non-local interactions without separate training on short and long disordered regions.