When does accounting for gene-environment interactions improve complex trait prediction? A case study with Drosophila...

Area of Science:

• Genetics
• Evolutionary Biology
• Quantitative Biology

Background:

• Gene-environment interactions (G×E) contribute to complex trait variation across species.
• Previous studies show G×E improves prediction accuracy in agriculture but not in human genetics.
• Understanding the discrepancy in G×E's predictive power is crucial for complex trait analysis.

Purpose of the Study:

• Investigate the reasons for contradictory findings regarding G×E in agricultural species versus human studies.
• Determine the specific conditions under which G×E improves prediction accuracy.
• Clarify the role of G×E in explaining trait variance and its implications for predictive modeling.

Main Methods:

• Utilized Drosophila melanogaster lifespan data with known G×E evidence.
• Developed three cross-validation scenarios simulating agricultural and human study population structures.
• Compared statistical models with and without G×E terms to assess prediction accuracy.

Main Results:

• Gene-environment interactions explained 8% of the variance in lifespan.
• Improved prediction accuracy with G×E was observed only when genotypes were shared between reference and test populations.
• This shared genotype scenario is typical in agricultural studies but absent in human studies with unrelated individuals.

Conclusions:

• G×E's contribution to prediction accuracy is contingent on the population structure of the reference and test datasets.
• The lack of improvement in human studies is likely due to the absence of shared genotypes between populations.
• This study provides a framework for understanding when to incorporate G×E in predictive models for complex traits.