Data coherence over data volume drives generalisable genome-based prediction of microbial carbon utilisation
- 1Oak Ridge National Laboratory
Abstract
Microbial carbon utilisation shapes ecological niches, community assembly, and nutrient cycling, yet remains difficult to predict from genomes despite well-characterised pathways. Genome-based machine-learning models often generalise poorly across datasets, a limitation commonly attributed to small or taxonomically biased training sets. We integrated binary growth phenotypes for 819 strains across 240 carbon sources from four datasets, focusing on 15 shared sources. Cross-dataset evaluation reduced balanced accuracy from 0.86 (within datasets) to 0.62, and restricting evaluation to in-clade relatives recovered only 0.03 of this loss. Because phylogeny alone could not explain the transfer gap, we restricted training to samples with coherent genotype-phenotype relationships, defined by agreement between experimental labels and annotated pathways. Concordance filtering doubled the number of carbon sources with reproducible genomic signals across datasets (from 6 to 12 of 15), often recovering known pathway genes, while matched random subsets remained at the unfiltered level. The smaller concordant set also approached the cross-dataset performance of models augmented with up to 1800 BacDive genomes per phenotype, demonstrating greater efficiency per sample. Concordance-trained models remained bounded specialists, but separated growers from non-growers among genomes the pathway rules called negative, mostly generalists with broad-specificity transporters. Model confidence made these bounds actionable, since abstention improved accuracy where models transferred and retraining on newly labelled low-confidence genomes gave the largest gains for weaker phenotypes. Collectively, these findings recast generalisation as a problem of label-mechanism agreement and demonstrate how genome-based predictions can be deployed within explicit applicability domains and improved through targeted phenotyping.
Related
- Beyond Better Models: Barriers and a Path Forward in Generalisable TCR Antigen Specificity Prediction
- City-specific or generalisable resistances? An urban animal connectivity model performs better when parameterised from two cities
- Proactive early invisible spoilage-risk prediction through uncertainty-quantified temporal microbial growth intelligence
- Peer‑led bioethics education in clinical rotations: Three generalisable lessons from a CFIR‑informed implementation case study
- In silico prediction of metabolic trait robustness in microbial cells
BibTeX
@misc{kishore2026data,
title = {{Data coherence over data volume drives generalisable genome-based prediction of microbial carbon utilisation}},
author = {D. Kishore and P. Ranjan and C. Neely and M. Cashman and Z. Sakkaff and W. J. Riehl and M. P. Joachimiak and J. N. Edirisinghe and J. P. Faria and P. Weisenhorn and M. B. Cohen and D. A. Pelletier and R. W. Cottingham and C. S. Henry and M. J. Doktycz and A. P. Arkin and P. S. Dehal},
howpublished = {bioRxiv},
year = {2026},
month = aug,
doi = {10.64898/2026.08.06.743333},
url = {https://www.biorxiv.org/content/10.64898/2026.08.06.743333v2},
note = {CC BY},
}
RIS
TY - UNPB
TI - Data coherence over data volume drives generalisable genome-based prediction of microbial carbon utilisation
AU - D. Kishore
AU - P. Ranjan
AU - C. Neely
AU - M. Cashman
AU - Z. Sakkaff
AU - W. J. Riehl
AU - M. P. Joachimiak
AU - J. N. Edirisinghe
AU - J. P. Faria
AU - P. Weisenhorn
AU - M. B. Cohen
AU - D. A. Pelletier
AU - R. W. Cottingham
AU - C. S. Henry
AU - M. J. Doktycz
AU - A. P. Arkin
AU - P. S. Dehal
PY - 2026
DA - 2026/08/06
PB - bioRxiv
DO - 10.64898/2026.08.06.743333
UR - https://www.biorxiv.org/content/10.64898/2026.08.06.743333v2
AB - Microbial carbon utilisation shapes ecological niches, community assembly, and nutrient cycling, yet remains difficult to predict from genomes despite well-characterised pathways. Genome-based machine-learning models often generalise poorly across datasets, a limitation commonly attributed to small or taxonomically biased training sets. We integrated binary growth phenotypes for 819 strains across 240 carbon sources from four datasets, focusing on 15 shared sources. Cross-dataset evaluation reduced balanced accuracy from 0.86 (within datasets) to 0.62, and restricting evaluation to in-clade relatives recovered only 0.03 of this loss. Because phylogeny alone could not explain the transfer gap, we restricted training to samples with coherent genotype-phenotype relationships, defined by agreement between experimental labels and annotated pathways. Concordance filtering doubled the number of carbon sources with reproducible genomic signals across datasets (from 6 to 12 of 15), often recovering known pathway genes, while matched random subsets remained at the unfiltered level. The smaller concordant set also approached the cross-dataset performance of models augmented with up to 1800 BacDive genomes per phenotype, demonstrating greater efficiency per sample. Concordance-trained models remained bounded specialists, but separated growers from non-growers among genomes the pathway rules called negative, mostly generalists with broad-specificity transporters. Model confidence made these bounds actionable, since abstention improved accuracy where models transferred and retraining on newly labelled low-confidence genomes gave the largest gains for weaker phenotypes. Collectively, these findings recast generalisation as a problem of label-mechanism agreement and demonstrate how genome-based predictions can be deployed within explicit applicability domains and improved through targeted phenotyping.
ER -