TY - JOUR T1 - IDENTIFICATION OF NOVEL GENES ASSOCIATED WITH PLATELET ACTIVATION SIGNALLING PATHWAYS IN HIGH DIMENSIONAL DATA USING AN ALTERNATIVE REGRESSION APPROACH JF - Heart JO - Heart SP - A2 LP - A2 DO - 10.1136/heartjnl-2014-306916.5 VL - 100 IS - Suppl 4 AU - BR Salehe AU - LJ McGuffin AU - CI Jones AU - G Di Fatta Y1 - 2014/12/01 UR - http://heart.bmj.com/content/100/Suppl_4/A2.2.abstract N2 - Platelet activation involves different signalling pathways in the underlying thrombus formation process. These pathways are the result of platelet responses to agonists' activation. Previous analyses involving four pathways (P-selectin in response to adenosine diphosphate (ADP), P-selectin in response to cross-linked polypeptide (CRP), fibrinogen binding stimulated with ADP and fibrinogen binding stimulated with CRP) revealed genomic associations regulating these pathways. These analyses were performed on single nucleotide polymorphisms data (SNPs) in which the underlying characteristic of these data normally contains small number of observations (N) and large number of variables or features (p). However, the methodologies used in analysing these genomic data involved linear models using stepwise regression. We argue that this approach deemed to be sub optimal for linear modelling analyses. We propose an alternative approach using more rob ust methods such as ridge regression and LASSO that would produce previously unknown novel SNPs describing their effects on the four pathways in the available large pool of SNPs. Methodology The genome-wide association (GWA) data containing 1554 single nucleotide polymorphisms (SNPs) for 512 individuals describing their effects on four signalling pathways were previously analysed statistically using stepwise regression[1] which is sub optimal[2]. We statistically re-analysed these data using both stepwise regression and shrinkage approaches. Results Several of the SNPs and their associated genes identified using our new stepwise approach were not previously selected, though are now found to be significant. Conclusion We propose shrinkage approach for linear models using ridge regression and LASSO for statistical analysis of genomic data with large p and small N. ER -