Overview.

We will request funding to employ a post-doctoral RA to carry out these analyses. The RA will be based at the PI's institution (Hong Kong Polytechnic University) and supervised by the PI and co-I's.

We seek approval from the ALSPAC Executive Committee to run the proposed analyses on the blue crystal computing cluster. The analyses are very computationally intensive (fitting a mixed model for each of millions of SNPs). We will include in the grant any funds specified by the Executive Committee required to cover this computing time as well as other appropriate costs incurred by ALSPAC. (NB. The longitudinal GWAS for BMI, which examined ~3 million SNPs, required approximately 8648 days of computing time to complete).

Aim.

To identify genetic variants influencing refractive error trajectory (either in the general population or in subjects classified as spending a high/low amount of time reading or outdoors).

Methods.

The refractive error and genotyping data has already undergone QC filtering for use in a closely related project that has received Exec approval: B1352 'CREAM (Consortium for Refractive Error and Myopia) 1000-genomes meta-analysis'.

Preliminary analysis of refractive error trajectories using linear mixed models are very promising (for example, as part of the above amendment work, we have been able to identify known myopia SNPs that (a) act early in childhood and retain stable effects during childhood, (b) exert a progressively increasing effect during childhood, and (c) interact with time spent outdoors or time spent reading.

However, fitting the model for each SNP currently takes ~1 minute. We propose to test simplified models and examine the robustness of fixed effects estimates, in order to minimse the time required to fit each model. Once optimised, we will conduct three longitudinal GWAS analyses for three models:

Model 1. Test for SNP main effects and 'SNP x age' interaction.

Model 2. As model 1, plus a test for 'SNP x nearwork' interaction.

Model 3. As model 1, plus a test for 'SNP x time outdoors' interaction.

We will aim to examine the ~4.5 million very high confidence imputed (quality metric RSQR greater than 0.8) very common variants (minor allele frequency, MAF >=0.1) SNPs in the 1000G-imputed dataset.