Abstract:
|
We did heritability (h2) estimates on drug response with SNP data. If we define the true portion of available SNPs on variance explained (VE) as h2M and then the SNPs potentially can explain all the genetic variation in the trait (h2M < = h2 ). The VE by genome-wide significant (GWS) SNPs (h2GWS) may satisfy h2GWS < h2M < =h2. However, in model building, we also consider: 1). use only the SNPs from target genes; 2). use the SNPs with less linkage disequilibrium among top SNPs, and 3). use the SNPs with MAF to be > 0.01. Subsequently, only SNPs with non-zero parameter estimates will be used in a 'molecular signature (MS)'. Then VE by MS may be even smaller than h2GWS i.e. h2MS < h2GWS < h2M < = h2. Two sets of estimates on real data are reported using GCTA method (Yang et al. 2011). 1). Impact of the SNP selection on h2 estimates, we found biased results if the same data were used on 3000 top SNPs; 2). Estimates of h2 on whole genome (0.99± 0.33, p< 0.0001) and on MS to predict drug response from discovery (0.009±0.002, p=0.34 with PLA arm and 0.08±0.05, P=0.002 with Trt arm) and validation (0.013±0.015, p=0.13, with Trt arm) data.
|