S-Pred

S-Pred: Protein Secondary Structure and Disorder Prediction Tool

S-Pred predicts eight-state secondary structures (SS8), accessible surface areas (ASAs), and intrinsically disordered regions (IDRs) from amino acid sequences using multiple sequence alignment (MSA)-derived features.

Key Features:

• Predictive Capabilities: Predicts SS8, ASAs, and IDRs directly from amino acid sequences.
• MSA-Based Input: Utilizes multiple sequence alignment (MSA) to capture evolutionary information for structural prediction.
• MSA Transformer: Employs an attention-based protein language model to extract feature representations from MSA inputs.
• Long Short-Term Memory (LSTM): Processes MSA Transformer-derived features to generate final structural and disorder predictions.
• Performance Metrics: Achieves ~76% SS8 accuracy, Pearson correlation coefficient of 0.84 for ASAs, and F1-score of 0.514 for IDR prediction.

Scientific Applications:

• Protein Function Prediction: Infers functional properties from predicted SS8, ASAs, and IDRs.
• Structural Modeling: Supports three-dimensional protein modeling using predicted secondary structures and surface accessibility.

Methodology:

S-Pred processes multiple sequence alignments using an attention-based MSA Transformer to generate sequence feature embeddings, which are subsequently analyzed by a Long Short-Term Memory (LSTM) network to predict eight-state secondary structures, accessible surface areas, and intrinsically disordered regions.