Skip to content

Coefficient of Determination (R-squared)

Summary

The Coefficient of Determination ($R^2$), commonly referred to as R-squared, is a key metric used in genetic prediction models of multi-omic traits to quantify predictive performance. It represents the proportion of variance in the target molecular trait (e.g., gene expression, protein abundance, metabolite level) that is explained by the genetic variants (SNPs) included in the score.

Application in Multi-omic Imputation

  • Model Evaluation: In platforms like OmicsPred, $R^2$ is calculated in held-out training data (using cross-validation) or external validation cohorts to assess model reliability.
  • Filtering Thresholds: Imputation workflows (such as TWAS or PWAS) often apply filters to exclude models with low predictive power (e.g., $R^2 < 0.01$ or $R^2 < 0.05$) to reduce false positives and improve statistical power.

Citations

  • Foguet, C., et al. (2026). OmicsPred as a centralised resource for genetic prediction of multi-omic traits. medRxiv preprint. DOI: 10.64898/2026.05.15.26353298.