Domain Composition Analysis

Domain-Composition Analysis

Summary

Analyze the Pfam domain and subPfam composition of genes within Biosynthetic Gene Clusters to identify and characterize domain patterns that distinguish sub-clusters. This skill enables tokenization and statistical profiling of BGC gene content to support sub-cluster detection and functional annotation.

When to use

Apply this skill when you have a set of Biosynthetic Gene Clusters in GenBank format and need to understand the domain architecture of constituent genes, either to detect statistically significant domain co-occurrence patterns within BGCs, to filter redundant BGCs by domain similarity, or to link detected sub-clusters to specific natural product substructures.

When NOT to use

• Input BGCs are already pre-filtered for domain redundancy or have been extensively curated; re-tokenization and re-filtering may introduce unnecessary computational overhead.
• Your goal is general BGC annotation or homology search rather than detection of statistically significant intra-cluster domain patterns or natural product sub-cluster linking.
• Pfam or subPfam annotations are unavailable or unreliable for the gene set; domain-based tokenization will not produce valid input for PRESTO-STAT.

Inputs

• Biosynthetic Gene Clusters in GenBank format
• Pfam domain annotations per gene
• subPfam annotations (optional, for increased resolution)

Outputs

• Tokenised BGC representations (gene × domain/subPfam vectors)
• Redundancy-filtered tokenised BGCs (via Adjacency Index similarity network)
• Detected sub-clusters with constituent genes and domain compositions
• Genomic positions and structural annotations of sub-clusters
• Structured sub-cluster output (linked to natural product substructures where available)

How to apply

Tokenize each gene in the BGC by representing it as a vector of its Pfam domains, using subPfams to increase resolution where available. Compute the Adjacency Index of domains as a distance metric to quantify similarity between BGC domain compositions. Use this similarity network to filter the tokenised BGC set for redundancy, removing highly similar BGCs. For remaining non-redundant BGCs, apply the PRESTO-STAT statistical algorithm (based on Del Carratore et al. 2019) to detect statistically significant co-occurring domain patterns across the tokenised sequences. Extract and record detected sub-clusters with their constituent genes, full domain compositions, and genomic positions for downstream annotation and validation.

Related tools

• iPRESTO (Command-line tool that orchestrates tokenization of BGCs using Pfam domains and subPfams, redundancy filtering via Adjacency Index similarity networks, and application of PRESTO-STAT and PRESTO-TOP statistical methods for sub-cluster detection and linking to natural product substructures.)

Evaluation signals

• Tokenised BGC representations must contain valid Pfam and subPfam domain identifiers for every gene; check that no gene is represented as an empty token vector.
• Adjacency Index similarity values between BGC pairs must fall in the range [0, 1]; BGCs with similarity above a specified threshold (e.g., 0.9) should be marked as redundant and one retained per cluster.
• PRESTO-STAT sub-clusters must have statistically significant co-occurrence p-values (typically p < 0.05 or adjusted p < 0.01); verify that p-values are computed and reported for each detected sub-cluster.
• Detected sub-clusters must have genomic positions that fall within the span of their parent BGC; validate that all gene positions are contiguous or within acceptable gap thresholds.
• Sub-clusters linked to natural product substructures should have interpretable domain compositions consistent with known biosynthetic pathways (e.g., polyketide synthase domains for PKS-derived products).

Limitations

• PRESTO-STAT relies on statistical power from sufficient redundancy-filtered BGC samples; small datasets may yield false negatives or unreliable co-occurrence estimates.
• Pfam and subPfam annotations are dependent on the completeness and curation of the reference database; novel or divergent domains may be missed or misclassified.
• Adjacency Index similarity is a coarse metric for redundancy filtering and may over-cluster functionally distinct BGCs with similar domain inventories but different arrangements.
• Sub-cluster detection is sensitive to tokenization granularity (i.e., choice of subPfams); different resolutions may yield overlapping or conflicting sub-cluster calls.
• Linking to natural product substructures requires a separate trained model or knowledge base; this skill produces sub-clusters but does not guarantee valid structure prediction.

Evidence

• [intro] BGCs are tokenised by representing each gene as a combination of its Pfam domains, where subPfams are used to increase resolution: "BGCs are tokenised by representing each gene as a combination of its Pfam domains, where subPfams are used to increase resolution"
• [intro] Tokenised BGCs are filtered for redundancy using similarity network with Adjacency Index of domains as distance metric: "Tokenised BGCs are filtered for redundancy using similarity network with an Adjacency Index of domains as a distance metric"
• [intro] PRESTO-STAT is based on statistical algorithm from Del Carratore et al. 2019 for detecting gene sub-clusters: "PRESTO-STAT, which is based on the statistical algorithm from Del Carratore et al. (2019)"
• [other] Sub-clusters are extracted with constituent genes, domain compositions, and genomic positions: "Extract and record detected sub-clusters with their constituent genes, domain compositions, and genomic positions"
• [intro] Sub-clusters found with iPRESTO can be linked to Natural Product substructures: "The sub-clusters found with iPRESTO can then be linked to Natural Product substructures"