Combining data types to predict lodgepole pine across environments
Source published
A bioRxiv preprint tests whether genomic, climate, soil and metabolite information improves lodgepole pine predictions when an environment is left out of model training.
AI-generated conceptual illustration; not documentary evidence.
Tree performance can vary between environments, complicating predictions from one set of trials to another. Researchers tested whether genomic, environmental and metabolomic data could improve predictions for lodgepole pine across the environments in their study.
Predicting performance across environments
Genotypes may not perform identically across environments, a pattern known as genotype-by-environment interaction. When such interactions occur, models trained using some environments may be less reliable for predicting performance in others. A study of lodgepole pine examined whether adding environmental and metabolomic information to genomic prediction models could help address this challenge.
The work is described in a bioRxiv preprint by E. P. Cappa and colleagues. The evidence available here is the abstract, so details beyond its description cannot be independently assessed from the supplied material.
What the researchers did
The study examined 12 traits measured on 11,124 trees across four lodgepole pine progeny trials. A subset of 1,490 trees was genotyped at 25,099 single-nucleotide polymorphisms (SNPs). The models also incorporated enviromic information—climatic and soil descriptors—and metabolomic information from targeted monoterpene measurements.
The researchers used single-step genomic best linear unbiased prediction (ssGBLUP) reaction-norm models. In broad terms, the reaction-norm approach modeled genotype-specific responses across environments. The abstract says the models included environmental covariates and both linear and nonlinear kernels, including Gaussian and Arc-Cosine formulations.
Predictive performance was assessed with leave-one-environment-out cross-validation: an environment was left out during training and used to evaluate predictions. This tests performance across the environments represented in the study; the abstract does not establish how well the approach would transfer to other regions or future conditions.
Findings reported in the abstract
The abstract reports substantial genotype-by-environment interaction for early-growth traits and drought-related decline. Additive genetic correlations among environments were frequently below 0.65. The abstract does not provide further detail on how these correlations varied by trait or environment.
Adding environmental covariates reduced variance attributed to site effects. The authors report that models incorporating genotype-by-environment (GE) and genotype-by-environmental-covariate (GW) interactions improved performance relative to baseline models. The abstract gives maximum reported changes of up to 29.58% more explained phenotypic variance and up to 3.13% less residual variance. These are upper-end results, not estimates of typical gains across traits.
On average across traits, GE effects explained 2.31% of phenotypic variation, compared with 0.60% for GW effects. Predictive ability across models ranged from 0.688 to 0.701. The abstract reports gains of up to 7.85% when nonlinear kernels were combined with metabolomic covariates, particularly for traits with stronger GE. Monoterpene profiles yielded predictive performance comparable to climate and soil descriptors in the reported analyses.
What the evidence cannot establish
The findings concern the evaluated lodgepole pine data and cross-validation setup. They do not show that the same improvements would occur in other populations, traits, regions or trial designs. Nor do they demonstrate that any data type causes improved tree performance or establishes greater climate resilience.
Because this account is based on the abstract rather than the full paper, it cannot assess the detailed methods, trait-by-trait results, uncertainty around estimates or supplementary analyses. The source is a bioRxiv preprint; the supplied evidence does not establish its review status. Its results should therefore be read as reported findings, not as settled evidence.
The study presents a way to test whether combining genomic, environmental and metabolomic information can improve prediction across the environments examined. Further evidence would be needed to determine how broadly the reported results apply.
lodgepole pine
genomic prediction
genotype-by-environment interaction
reaction norm models
environmental covariates
monoterpenes
Related reading
Automatically matched by shared topic keywords. These links do not imply that the sources agree or establish a consensus.