Advancing My Training in Genomic Prediction for Forest Tree Breeding: Insights from the Genome-Wide Prediction of Complex Traits Workshop

As part of my postdoctoral training in the Momentum Program at CSIC, I recently completed the Genome-Wide Prediction of Complex Traits workshop (March 24–28, 2025) organized by Physalia courses and instructed by the Dr Oscar González Regio (INIA-CSIC) and Dr. María Evangelina López de Maturana. This intensive five-day program provided a deep dive into genomic prediction, equipping me with both theoretical and practical knowledge on applying statistical and machine learning models to genomic and phenotypic data. The course was held online, allowing for international participation and engagement with a diverse group of researchers.

Throughout the course, I gained a comprehensive understanding of the fundamental steps involved in genomic prediction. We started with an introduction to genome-wide prediction in human genetics, animal breeding, and plant breeding, revisiting key concepts of quantitative genetics and linear mixed models. From there, we explored the challenges posed by high-dimensional datasets, particularly in large p, small n scenarios, and compared traditional pedigree-based resemblance with genomic-based approaches.

One of the most valuable aspects for me was delving into different statistical methodologies for genomic prediction. We covered shrinkage estimation techniques such as GBLUP and kernel-based regression models, as well as Bayesian methods for SNP regression. Additionally, we explored machine learning approaches, evaluating their potential for genomic prediction and how they compare to traditional statistical methods.

Another crucial component of the course was understanding predictive ability metrics. We worked with MSE, Pearson and Spearman correlations, AUC-ROC curves, and cross-validation strategies to assess model performance. These techniques are essential for ensuring that genomic prediction models are not only statistically sound but also practically useful in breeding programs and medical applications.

The course concluded with a hands-on workshop on genome-enabled prediction, where we applied the concepts learned throughout the week. Working with real datasets, I implemented genomic prediction pipelines using R, Linux command line tools, and specialized software. This hands-on experience solidified my understanding of the entire workflow, from data preprocessing and imputation to model evaluation.

Beyond the technical skills, one of the most rewarding aspects was engaging with the instructors and other participants through video discussions and a dedicated Slack channel. The interactive format allowed for meaningful discussions on methodological choices and their applications in different biological fields.

Completing this course has significantly strengthened my expertise in genomic prediction, directly benefiting my research on integrating multi-omics data for predicting complex traits in forest trees. The knowledge gained will be instrumental in refining my models for traits such as productivity, disease resistance, and climate adaptation. I am excited to apply these methodologies to my ongoing work and to collaborate with peers who share an interest in advancing genomic prediction.