Enzyme Substrate Classification: SDRs and SAM-MTases with AlphaFold2 Structures

Adrián Jinich from Cornell University compiled sequence, structure, and substrate data for 358 short-chain dehydrogenase/reductases (SDRs) and 953 S-adenosylmethionine dependent methyltransferases (SAM-MTases). Protein sequences were sourced from UniProtKB with an annotation score above 4-out-of-5, and three-dimensional structures were predicted using AlphaFold2 via ColabFold. Substrate classification labels were derived from UniProtKB annotations using both automated clustering based on molecular fingerprints and manual classification tasks.

Use Cases

• Training machine learning models for enzyme substrate classification based on sequence and structural features.
• Benchmarking protein structure prediction tools using the included AlphaFold2-generated PDB files.
• Analyzing relationships between protein family, structure, and substrate specificity for SDRs and SAM-MTases.
• Developing substructure search methods for substrate annotation using the provided SMARTS strings.

Strengths

• Includes 358 SDR and 953 SAM-MTase enzyme entries, providing a substantial sample size for each family.
• Integrates multiple data types: amino acid sequences, predicted 3D structures from AlphaFold2, and substrate classification labels.
• Data quality was filtered using a UniProt annotation score threshold of above 4-out-of-5.
• Provides both automated substrate clustering (9 clusters for SDRs, 13 for SAM-MTases) and manually defined classification tasks.

Limitations

• Row count and file size are unknown, which may limit suitability assessment for large-scale projects.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

UniProt Knowledgebase (UniProtKB), AlphaFold2 structure predictor via ColabFold.

Collection Method

Sequences obtained via InterPro family queries (IPR002347, IPR029063), filtered by annotation score and deduplicated. Structures predicted with AlphaFold2. Substrate labels derived from UniProtKB annotations.