CSM

CSM predicts enzyme activity (including enzyme commission (EC) numbers) and classifies protein structures (SCOP class, superfamily, family, fold) by converting residue–residue distance patterns at multiple cutoffs into feature vectors and applying singular value decomposition for classification.

Key Features:

• Structure-Based Methodology: Analyzes spatial relationships between amino acid residues using distance cutoffs to produce residue–residue distance matrices.
• Feature Vector Generation: Converts residue–residue distance matrices into feature vectors that encapsulate distance pattern information for downstream analysis.
• Dimensionality Reduction via Singular Value Decomposition (SVD): Applies singular value decomposition to reduce dimensionality and mitigate noise prior to classification.
• High Precision and Recall: Achieves reported precision up to 99% and recall up to 95% on datasets including manually curated protein superfamilies and extensive enzyme commission (EC) number datasets.
• Versatile Application in Structural Classification: Assigns SCOP class, superfamily, family, and fold to protein domains with high precision and recall on comprehensive domain sets from the latest SCOP release.
• Improved Recall with Consistent Precision: Demonstrates enhanced recall compared to previous studies while maintaining comparable levels of precision.

Scientific Applications:

• Automatic Annotation: Predicts enzyme activity (including EC numbers) and structural classifications to support automated annotation of protein datasets.
• Structural Genomics: Classifies proteins at class, superfamily, family, and fold levels to aid cataloging and analysis in structural genomics.
• Enzyme Research: Aids identification of potential enzymatic functions of uncharacterized proteins to accelerate studies of metabolic pathways and drug development.

Methodology:

Constructs residue–residue distance matrices at multiple cutoff thresholds from protein structures, converts these matrices into feature vectors, applies singular value decomposition (SVD) for dimensionality reduction, and uses the reduced vectors for classification and enzyme-activity prediction.