Fig. 1.

A penalized regression-based approach to identify hematopoietic regulatory loci and genes. a Schematic outline of our approach. We generated a penalized regression-based predictive scoring algorithm based on platelet trait GWAS loci and applied the resultant scoring algorithm genome-wide to predict causal variants and genes. We validated this model computationally and through validation of TPM1 function in induced pluripotent stem cells (iPSCs). b To generate a penalized regression model, 580 platelet trait GWAS SNPs [6] and matched control SNPs (~ 100 per GWAS SNP [25]) were analyzed for overlap with 860 chromatin features (e.g., histone marks and transcription factor binding sites). c Penalized regression (LASSO [23]) analysis identified 38 chromatin features from the indicated cell types that best discriminated GWAS SNPs, after controlling for background features (distance to nearest gene, number of SNPs in linkage disequilibrium, and minor allele frequency). Bar heights are LASSO coefficients, indicating the relative importance of each feature. MK, primary megakaryocytes; Ery, peripheral blood derived erythroblasts; MK/Ery, K562 cells; Lymphoblast, GM12878 or GM12891