EnzRank

EnzRank ranks enzymes using a convolutional neural network to prioritize candidates for re-engineering toward novel substrate activities in retro-biosynthetic and pathway design.

Key Features:

• Convolutional Neural Network: EnzRank employs a convolutional neural network trained on 11,800 known active enzyme–substrate pairs sourced from the BRENDA database.
• Negative Sample Generation: Negative samples are generated by scrambling known enzyme–substrate pairs to complement the positive training set.
• Substrate Dissimilarity Assessment: The tool uses Tanimoto similarity scores to assess dissimilarity between an enzyme's native substrate and candidate molecules.
• Input Data Formats: Model inputs include SMILES strings for substrates and enzyme sequences.
• Performance Metrics and Validation: Performance is reported as average recovery rates of 80.72% for positive pairs and 73.08% for negative pairs, assessed with a 10-fold holdout cross-validation.

Scientific Applications:

• De Novo Pathway Design: Prioritizes enzymes as starting points for re-engineering to construct novel biosynthetic pathways.
• Retro-Biosynthesis: Predicts potential secondary activities of enzymes to identify alternative production routes beyond cataloged enzymatic activities.
• Protein Engineering: Ranks enzymes by suitability for directed evolution or de novo design in synthetic biology and metabolic engineering.

Methodology:

Training a CNN on 11,800 positive enzyme–substrate pairs from BRENDA with scrambled negative samples, computing Tanimoto similarity scores for substrate dissimilarity, and evaluating performance using 10-fold holdout cross-validation.