Comparative analysis of genomic prediction approaches for multiple time-resolved traits in maize

Area of Science:

• Plant genetics
• Genomic prediction
• Agricultural science

Background:

• Genomic prediction is crucial for crop breeding and precision agriculture.
• Current methods often predict traits independently, neglecting developmental dynamics.
• High-throughput phenotyping provides rich time-series data for multiple traits.

Purpose of the Study:

• To compare MegaLMM and dynamicGP for multi-trait genomic prediction using temporal data.
• To evaluate predictive performance for both single time points and trait dynamics.
• To analyze factors influencing the prediction of trait developmental trajectories.

Main Methods:

• Utilized time-series data for 50 traits in a maize MAGIC population.
• Compared MegaLMM and dynamicGP, including hybrid variants.
• Assessed performance using snapshot and longitudinal accuracy (PCC and MSE).

Main Results:

• MegaLMM demonstrated superior snapshot and longitudinal Pearson correlation coefficient (PCC) compared to dynamicGP.
• No significant difference in snapshot mean squared error (MSE) was observed between the two models.
• DynamicGP uniquely predicted multiple traits beyond the training time points.

Conclusions:

• MegaLMM offers improved accuracy for predicting multiple traits over time.
• DynamicGP provides valuable forecasting capabilities beyond observed data.
• Further research is needed to understand factors affecting the prediction of trait dynamics from genetic markers.