Machine learning accurately predicts the multivariate performance phenotype from morphology in lizards.

Area of Science:

• Evolutionary biology
• Quantitative genetics
• Computational biology

Background:

• Completing the genotype-to-phenotype map requires comprehensive organismal phenotype measurement.
• Large-scale phenotyping is challenging, leading to missing data that hinders comparative analyses and evolutionary trend assessments.
• Predicting multivariate performance traits from morphology is particularly difficult due to logistical constraints.

Purpose of the Study:

• To develop a machine learning model for accurately estimating multivariate performance data solely from morphological measurements.
• To address the challenge of missing performance data in large-scale phenotyping studies.
• To improve the prediction of performance traits crucial for understanding evolutionary trends.

Main Methods:

• A machine learning model was developed and trained on a dataset of performance and morphology data from 68 lizard species.
• A stacked model architecture was utilized for enhanced predictive accuracy.
• The model was evaluated on its ability to predict missing performance data from simple morphological measures.

Main Results:

• The developed machine learning model accurately estimates multivariate performance data from morphology alone at the individual level.
• The model performed exceptionally well, even for traits with over 90% missing values.
• Incorporating phylogenetic information did not improve the model's predictive fit, indicating phenotypic data alone was sufficient.

Conclusions:

• This machine learning approach effectively predicts performance phenotypes from morphology, overcoming challenges of missing data.
• The method enhances our understanding of performance evolution and can integrate performance data into future phenomics research.
• Phylogenetic information is not essential for accurate performance prediction when sufficient morphological data is available.