Authors
Wu, T.
Abstract
Motivation: Genome-scale metabolic models contain thousands of reactions, yet the pathway maps used to interpret them are still largely drawn by hand and cover only a small fraction of each model. Existing automatic layout methods mainly evaluate geometric appearance, such as edge crossings and spacing. However, a map can look clean while representing a reaction through the wrong metabolites, placing it on the wrong biological pathway. Automatic metabolic mapping therefore requires deciding not only where to draw reactions, but also which biological connection each reaction should represent. Results: mytool addresses both problems by representing each reaction with one primary substrate--product connection. The pair is selected according to conserved chemical overlap while preventing ubiquitous carriers and cofactors from dominating the pathway backbone. Across four genome-scale models, the selected pair agrees with curated KEGG pathway drawings for 92.6--96.1% of scored reactions. Removing the carrier constraint reduces agreement by 3.9--6.3 percentage points, while removing cofactor handling reduces it by 35.8--43.3 points. A connectivity-based alternative achieves similar recall only by drawing 1.67--3.47 connections per reaction, with 31.7--65.4% precision; mytool draws exactly one. Across 676 maps from eight organisms, mytool produces geometric layout quality competitive with Graphviz DOT while generating substantially more orthogonal structure than force-directed layouts (median axis-aligned edge fraction 0.615 versus 0.000--0.019). The 10,600-reaction human Recon3D model is partitioned and drawn in 3.3 min. Every model in the BiGG database was drawn without manual intervention, producing 2621 maps spanning 240,398 reactions with no drawing failures; across the whole collection the median map places 0.988 of its edge length on an axis, at 0.056 crossings per edge. Together, these results show that genome-scale metabolic maps can be generated automatically while preserving the biological pathway structure needed for interpretation. Availability and implementation: Source code is available under CC BY 4.0 at https://github.com/forxhunter/MetaCarto github.com/forxhunter/MetaCarto. A collection of 2621 maps covering all 108 BiGG models, each as Escher JSON and SVG, is available at github.com/forxhunter/Awesome_visualization_Metabolic_Network and can be browsed interactively at forxhunter.github.io/escher}. [Zenodo DOI to be added.] Contact: [email protected] Supplementary information: Supplementary data are available at github.com/forxhunter/MetaCarto.
Preprint server:
bioRxiv
The authors list and abstract were imported from bioRxiv on 25 Sep 2026.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 5
- Comments 0