LSC Abstract – Prediction of COPD- and smoking status by network-based multi-'omics data fusion analysis

Our objective is to classify smoking- and COPD disease status by network-based multi-9omics data fusion analysis. We hypothesize that bridging and integration of multi-molecular level data will provide improved power for classification of smoking and COPD diagnosis. 7 9Omics data sets from the Karolinska COSMIC study were. Feature selections and a network-based multi-9omics data fusion and clustering of subjects analysis was performed. We found that the prediction power (Normalized Mutual Information, NMI between predicted clusters and known groups) was increased with the number of (complementary) 9omics data blocks used. The best predictor for all four groups (NMI=0.85) was achieved by integration of proteome, mRNA transcriptome data from BAL cells, combined with metabolomics and exosomal miRNA profiling from BAL fluid, including 276 significant COPD-influenced features (p The results indicate that there is both comprehensive and redundant information within multi-9omics datasets. The predictive power will be highly increased by integration of comprehensive multi-9omics data. Even for dominant factors smoking, the predictive power will be significantly improved by multi-9omics fusion than single 9omics. Further expansion of these methods to allow sub-phenotyping of at-risk smokers and COPD patients may lead to improved techniques for early diagnosis of COPD. The predictive power (NMI) is increased with the number of complementary 9OMICs.