SAMbinder

SAMbinder predicts S-adenosyl-L-methionine (SAM) binding residues in proteins from amino acid sequences to enable identification of SAM binding sites relevant to protein function and therapeutic targeting.

Key Features:

• Predictive Capability: SAMbinder uses machine learning to predict S-adenosyl-L-methionine (SAM) binding residues from primary protein sequences.
• Machine Learning Models: The method employs Random Forest algorithms trained on binary profile features and Position-Specific Scoring Matrix (PSSM) profiles to capture evolutionary information.
• Dataset Utilization: Training involved datasets comprising 2188 SAM-interacting residues balanced against non-interacting counterparts and evaluation on a more realistic dataset with an increased number of non-interacting residues.

Scientific Applications:

• Structural biology: Identification of SAM binding sites to inform structural interpretation of protein–ligand interactions.
• Protein function and interaction studies: Facilitation of understanding of protein function and SAM-related interaction mechanisms.
• Drug discovery: Support for designing therapeutics targeting SAM-associated diseases by locating potential binding residues.

Methodology:

Models were trained and tested on protein chains with less than 40% sequence similarity; features included binary profiles and PSSM profiles, with PSSM improving performance; evaluation used Matthews Correlation Coefficient (MCC) and Area Under the Receiver Operating Characteristic Curve (AUROC) — Random Forest achieved MCC 0.42 and AUROC 0.79 with binary profiles and MCC 0.61 and AUROC 0.89 with PSSM profiles; internal and external cross-validation were applied.