Letters and Biological Process: Linguistic Structure Encodes Hierarchical Depth in the Gene Ontology.

MOTIVATION: We asked whether the natural language used to name Gene Ontology (GO) Biological Process terms encodes hierarchical position in the GO directed acyclic graph.

RESULTS

Semantic representations of term names, combined with UMAP visualization and k-means clustering (k=20) recovered biologically coherent semantic groups whose depth distributions differed strongly (Kruskal-Wallis H=5244, p<10-300; η2=0.217). Term-name length increased with depth (Pearson r=0.34, Spearman ρ=0.37; slope ≈3.2 letters per level), and this positive scaling held under shortest-, longest-, and average-path depth definitions. Vocabulary entropy followed a bell-shaped diversification-specialization curve with an intermediate-depth peak (maximal at depth 4) that lies outside a Monte Carlo null envelope across most depths, and null-model tests supported a genuine linguistic-hierarchical association.

AVAILABILITY AND IMPLEMENTATION: The go_lingdepth Python package, regenerated result tables, and figure-generation scripts are available at https://github.com/Sadiya-ISU/Letters-and-Biological-Process and archived on Zenodo (DOI: 10.5281/zenodo.21501473).

SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.