Will Big Data Close the Missing Heritability Gap?

Area of Science:

• Genetics
• Bioinformatics
• Statistical Genomics

Background:

• Genome-wide association studies (GWAS) have identified genetic variants for many traits, but prediction accuracy remains low compared to heritability.
• The increasing availability of large biomedical datasets from biobanks presents an opportunity to enhance genomic prediction.
• A key question is whether 'big data' can bridge the gap between trait heritability and the variance explained by genomic predictors.

Purpose of the Study:

• To investigate the impact of sample size and model complexity on genomic prediction accuracy.
• To forecast the potential of large biobank datasets to improve prediction R-squared (R-sq.) for complex traits.
• To estimate the extent to which big data can reduce the gap between trait heritability and prediction accuracy.

Main Methods:

• Utilized Bayesian methods and a data analysis approach to model prediction R-sq. as a function of sample size and model complexity.
• Applied the methodology to UK Biobank interim release data, focusing on human height as a model trait.
• Trained models using 80,000 records and tested on 22,221 individuals to assess prediction R-sq.

Main Results:

• Achieved a prediction R-sq. of 0.24 in the testing set for human height.
• Demonstrated that prediction R-sq. increases with sample size, plateauing at different levels depending on the number of single nucleotide polymorphisms (SNPs) used.
• Estimated plateau R-sq. values ranged from 0.1 (500 SNPs) to 0.37 (50,000 SNPs).

Conclusions:

• Big data from modern biobanks will substantially reduce the gap between trait heritability and genomic prediction accuracy.
• Forecasted further improvements in prediction R-sq. with larger sample sizes.
• Anticipated that for complex traits, the gap between prediction R-sq. and heritability will not be entirely closed, even with big data.