TaxonBodyMassML: Taxonomy-informed prediction of body mass using gradient-boosted trees
Abstract
1. Body mass is a key ecological trait governing metabolic rates, life-history characteristics, interaction strengths, and community structure and dynamics. Despite its centrality to ecology and ecological modelling, direct body mass measurements exist for only a small fraction of all scientifically described species. 2. We present TaxonBodyMassML, a machine-learning tool to predict the body mass of heterotrophic taxa from their Linnaean taxonomy alone. Trained on 36,573 species-level body mass records compiled from 345 primary sources and databases, the underlying model combines learned entity embeddings of the kingdom-to-genus ranks with a gradient-boosted tree regressor to achieve R² = 0.91, RMSE = 0.56 log10 units, and MAE = 0.32 log10 units on a held-out test set of 3,657 species spanning approximately 22 orders of magnitude in body mass. Conformal prediction intervals, calibrated separately for each level of taxonomic resolution, provide user-selectable coverage without model retraining. 3. The tool accepts scientific names at any taxonomic level with tolerance for misspellings via fuzzy matching, optionally returns the measured mass for species in its database or a point estimate in grams with optional prediction intervals for these and all taxa, and requires no phylogenetic tree or correlated trait data. 4. TaxonBodyMassML is available as an R package, a Python package, and an open web interface (https://taxonbodymassml.github.io).
Related
- Ark-ADMET: endpoint-specific selection of molecular representations and hyperparameters for gradient-boosted ADMET models under the Therapeutics Data Commons benchmark protocol
- A Taxonomy-Informed Sparse DNA Foundation Model for Microbial Genomics
- Trait-informed metabarcoding captures tardigrade community trait shifts but not functional dispersion along a land-use gradient
- Comparative analysis of gradient boosting and random forest regressors for nutrient prediction in smart agriculture
- AlphaGenome informed machine learning for pathogenicity prediction of germline mutations
BibTeX
@misc{pasquantonio2026taxonbodymassml,
title = {{TaxonBodyMassML: Taxonomy-informed prediction of body mass using gradient-boosted trees}},
author = {Grant Mori Pasquantonio and Hailey Prater and Anish Chowdary Pendurti and Mark Novak},
howpublished = {EcoEvoRxiv},
year = {2026},
month = sep,
doi = {10.32942/x2bx01},
url = {https://ecoevorxiv.org/repository/view/15232/},
note = {CC BY-NC-SA 4.0},
}
RIS
TY - UNPB
TI - TaxonBodyMassML: Taxonomy-informed prediction of body mass using gradient-boosted trees
AU - Grant Mori Pasquantonio
AU - Hailey Prater
AU - Anish Chowdary Pendurti
AU - Mark Novak
PY - 2026
DA - 2026/09/28
PB - EcoEvoRxiv
DO - 10.32942/x2bx01
UR - https://ecoevorxiv.org/repository/view/15232/
AB - 1. Body mass is a key ecological trait governing metabolic rates, life-history characteristics, interaction strengths, and community structure and dynamics. Despite its centrality to ecology and ecological modelling, direct body mass measurements exist for only a small fraction of all scientifically described species. 2. We present TaxonBodyMassML, a machine-learning tool to predict the body mass of heterotrophic taxa from their Linnaean taxonomy alone. Trained on 36,573 species-level body mass records compiled from 345 primary sources and databases, the underlying model combines learned entity embeddings of the kingdom-to-genus ranks with a gradient-boosted tree regressor to achieve R² = 0.91, RMSE = 0.56 log10 units, and MAE = 0.32 log10 units on a held-out test set of 3,657 species spanning approximately 22 orders of magnitude in body mass. Conformal prediction intervals, calibrated separately for each level of taxonomic resolution, provide user-selectable coverage without model retraining. 3. The tool accepts scientific names at any taxonomic level with tolerance for misspellings via fuzzy matching, optionally returns the measured mass for species in its database or a point estimate in grams with optional prediction intervals for these and all taxa, and requires no phylogenetic tree or correlated trait data. 4. TaxonBodyMassML is available as an R package, a Python package, and an open web interface (https://taxonbodymassml.github.io).
ER -