Reaction Rule Coverage Analysis

reaction-rule-coverage-analysis

Summary

Empirically determine the scaling relationship between the number of generalized reaction rules applied and the percentage coverage of a reference reaction database (e.g., MetaCyc). This skill quantifies the trade-off between rule set size and reaction prediction coverage, enabling practitioners to select optimal rule subsets for their computational resources.

When to use

When you need to decide how many reaction rules to include in a Pickaxe expansion run and want to quantify the coverage penalty of using a smaller subset; when benchmarking rule importance or frequency rankings; or when optimizing rule selection for a constrained computational budget while maintaining acceptable reaction coverage.

When NOT to use

• You are working with reaction rules that are already curated or filtered to a specific biological context (e.g., organism-specific pathways); coverage analysis assumes a large, unbiased reference rule set.
• Your goal is to validate the correctness of individual reaction predictions, not to characterize aggregate coverage; this skill measures coverage breadth, not prediction accuracy.
• You have already selected a fixed rule subset based on prior knowledge or experimental evidence; coverage analysis is a planning/benchmarking tool, not a prediction validation step.

Inputs

• Full generalized reaction rule set (TSV format, ~1221 MetaCyc rules with SMARTS representations)
• Representative compound set (SMILES strings) from the reference database
• Reference reaction database (e.g., MetaCyc reactions in TSV or equivalent format)
• Rule ranking or frequency scores for subset selection

Outputs

• Coverage table (TSV format): rows of (number_of_rules, coverage_percent) pairs
• Scaling curve plot (optional): visualization of coverage vs. rule cardinality
• Validation report: empirical coverage values compared to expected benchmarks

How to apply

Load the full generalized rule set (e.g., ~1221 MetaCyc generalized rules from MINE-Database) and create ranked subsets at strategic cardinalities (e.g., 20, 100, 272, 500, 956, 1221 rules) ordered by rule importance or frequency. For each subset, apply the rules to a representative compound set from the reference database and compute the fraction of total reference reactions that can be represented by the selected rules. Aggregate coverage percentages as (number_of_rules, coverage_percent) pairs and tabulate results. Compare empirical scaling curve against reported benchmarks to validate that selecting N rules achieves the expected coverage; use these curves to identify the rule cardinality that meets your application's coverage threshold.

Related tools

• MINE-Database (Pickaxe) (Applies generalized reaction rules to compounds; provides the default MetaCyc generalized rule set and framework for iterative rule application) — https://github.com/tyo-nu/MINE-Database
• RDKit (Parses SMILES and SMARTS representations; enables rule-based molecular transformation and similarity calculations)
• Python (Orchestrates rule subset selection, iterative rule application, coverage computation, and tabulation)

Examples

# Load MINE-Database, select rule subsets [20, 100, 272, 500, 956, 1221], apply to MetaCyc compounds, and tabulate coverage:
from minedatabase.pickaxe import Pickaxe
rule_counts = [20, 100, 272, 500, 956, 1221]
coverage_table = []
for n_rules in rule_counts:
    rules_subset = load_rules_ranked('./data/metacyc_generalized_rules.tsv', n_top=n_rules)
    pickaxe = Pickaxe(rules=rules_subset, coreactants='./data/metacyc_coreactants.tsv')
    coverage_pct = compute_coverage(pickaxe, reference_db='MetaCyc')
    coverage_table.append((n_rules, coverage_pct))
with open('coverage_table.tsv', 'w') as f:
    f.write('num_rules\tcoverage_percent\n')
    for n, cov in coverage_table:
        f.write(f'{n}\t{cov}\n')

Evaluation signals

• Tabulated coverage values match or fall within expected ranges from published benchmarks (e.g., 20 rules → ~50% coverage, 272 rules → ~90% coverage, 1221 rules → 100% coverage for MetaCyc).
• Coverage percentages form a monotonically increasing curve with respect to rule cardinality; no rule subset should produce lower coverage than a larger subset.
• Sum of reactions covered by rule subset N ≥ sum covered by subset N−1; cumulative coverage is non-decreasing.
• Rule subsets are properly ranked by importance or frequency; top-ranked rules should dominate coverage gains at small cardinalities.
• All reference reactions are reachable by the full rule set (1221 rules achieve 100% coverage); this validates reference database completeness and rule-application logic.

Limitations

• Coverage scaling is specific to the reference reaction database (MetaCyc); coverage curves may differ for KEGG or other reaction sources.
• Rule ranking method (importance vs. frequency) affects the coverage curve; different ranking strategies will produce different cardinality-vs-coverage relationships.
• Coverage is measured on the reference database compounds only; coverage may not transfer to novel or user-supplied compound sets, which may trigger different reaction pathways.
• Computational cost scales with rule cardinality and compound set size; expansion of 10,000 KEGG compounds with 272 rules can yield 5 million compounds, limiting feasibility for large subsets.

Evidence

• [intro] Coverage relationship: "20 rules provide 50% coverage of MetaCyc reactions; 100 rules provide 78% coverage; 272 rules provide 90% coverage; 500 rules provide 95% coverage; 956 rules provide 99% coverage; 1221 rules provide"
• [intro] Rule set source and size: "A set of biological reaction rules and cofactors are provided by default. These consist of approximately 70,000 MetaCyc reactions condensed into generic rules"
• [intro] Pickaxe application workflow: "Pickaxe applies reaction rules, representing reaction transformation patterns, to a list of user-specified compounds in order to predict reactions"
• [readme] Iterative rule application: "Pickaxe creates a network expansion by applying these reaction rules iteratively to your starting set of compounds, going for as many generations as you specify"
• [intro] SMARTS rule representation: "Rules are generated using SMARTS which represent reactions in a string"
• [intro] Scalability example: "For example, expanding 10,000 compounds from KEGG with 272 rules from metacyc yields 5 million compounds"