NRPSpredictor2

NRPSpredictor2 predicts substrate specificity of bacterial and fungal non-ribosomal peptide synthetase (NRPS) adenylation domains to annotate and characterize biosynthetic gene clusters producing secondary metabolites such as antibiotics (e.g., vancomycin).

Key Features:

• SVM-based classification: Uses Support Vector Machines implemented with SVMlight to classify adenylation domain substrate specificity.
• Multi-level specificity hierarchy: Predicts adenylation domain specificity across four hierarchical levels from broad physicochemical properties to single amino acid substrates.
• Physico-chemical fingerprints: Derives physico-chemical fingerprints from protein sequence families and extracts residues surrounding substrate-binding positions.
• Normalized feature vectors: Encodes extracted residues into normalized feature vectors for SVM training and classification.
• Transductive SVMs: Applies transductive SVMs to incorporate unlabeled data and improve prediction reliability.
• Applicability domain modeling: Models an applicability domain to estimate whether new adenylation domains fall within the predictor's scope.
• Fungal NRPS support: Extends predictions to fungal NRPS adenylation domains, including substrates relevant to peptaibols and cephalosporins.
• Performance metrics: Reports F-measures >0.89 for three general levels, an average F-measure of 0.80 at the most detailed level, and 0.84 for fungal adenylation domains.
• UniProt coverage and comparison: Provides specificity predictions for an additional 18% of NRPS A domains detectable in UniProt and shows 70% agreement and <6% disagreement with existing methods.
• Novel specificity detection: Identifies sequences predicted to have completely new types of substrate specificity not characterized by current predictive methods.

Scientific Applications:

• Biosynthetic gene cluster annotation: Annotate biosynthetic gene clusters responsible for secondary metabolite production, including antibiotics such as vancomycin.
• Substrate assignment for pathway reconstruction: Assign substrates to NRPS adenylation domains across hierarchical specificity levels to support biosynthetic pathway reconstruction.
• Fungal natural product discovery: Predict fungal NRPS substrate specificities to support discovery of fungal natural products such as peptaibols and cephalosporins.
• Novel specificity identification: Flag candidate adenylation domains with novel predicted specificities for experimental characterization.

Methodology:

Extracts residues around substrate positions and derives physico-chemical fingerprints that are encoded into normalized feature vectors, then trains SVM classifiers using SVMlight across four hierarchical specificity levels and applies transductive SVMs while modeling an applicability domain.