511 Kullback-Leibler Divergence Model to Integrate Genetic and Genomic Information to Assess Drug Response for Psoriatic Patients

Psoriasis is an immune-mediated inflammatory and hyperproliferative skin condition affecting ∼2% of the US population, with a total annual cost of around 3 billion dollars. Despite the successes of drug development, there can be significant variation in treatment response, which can correlate with patients’ genetic variations and baseline skin genomic profiles. However, no study has integrated multiomic information to enhance drug response assessment, potentially because this data is rarely available from the same cohort, and current modeling techniques are limited in their ability to robustly integrate partially overlapping multi-view data. We seek to address the above limitations on a longitudinal RNA-seq cohort of 44 patients that received anti-TNF treatment with documented changes in PASI score, as well as an independent genetic cohort of 428 psoriatic patients with self-reported 5 level outcomes rating the drug prognosis. We used an advanced Kullback-Leibler divergence(KL) based integrative approach to model the multi-view information, leveraging information from genetics data to improve the drug response assessment from the genomics information. We used variant calling to identify common variations in the RNA-seq samples, and regularized regression (LASSO) to improve the identification of informative genetic and genomic markers for the complex trait sparsity structure. Compared with using genomics data alone, the integrative KL model reduced the 5-fold predictive mean squared error (MSE) by 3.8% from 2.61 to 2.51, improved the model R2 from 0.0193 to 0.459 and further identified >30 informative markers that can be used to enhance drug response prediction. Our method highlights the feasibility of using statistical techniques to analyze independent multi-modal biological data, thus providing a significant opportunity to integrate available information from different sources and improving the prognostic prediction accuracy.