A Multi-Molecular Fusion to Detect Transcriptomic Signature in Tissue-Specific Cancer
Suparna Saha1, Saurav Mallik2, Sanghamitra Bandyopadhyay31SyMeC Data Center, Indian Statistical Institute, Kolkata, India, 2Department of Environmental Epigenetics, Harvard T H Chan School of Public Health, Boston, USA, 3Machine Intelligence Unit, Indian Statistical Institute, Kolkata, India,
Analysis of multi-molecular interactions and detection of combinatorial transcriptomic signatures are emerging as important research topics in disease analytics. Currently, a combination of gene and miRNA expression proling in bioinformatic analysis enables us to comprehensively detect molecular changes in cancer and thereafter to identify integrated signatures and pathways that exist in the miRNA and gene interaction networks. Although many methodologies and applications have been suggested in recent literature, ecient techniques that can integrate the complex gene as well as miRNA expression proles, and identify the most relevant signatures are required. In this article, we presented a new framework of multimolecular data integration to identify combinatorial transcriptomic signatures through the strategy of unsupervised learning and target detection. Later, we evaluated their utility in survival analysis through a multivariate Cox regression study. We used a cervical cancer data repository to conduct our experiment. To construct the miRNA-mRNA interaction network, we selected the downregulated mRNAs that were negatively correlated with the upregulated miRNAs. Thereafter, we identied dense modules by using an unsupervised learning technique. The silhouette index value was computed for each cluster. By considering the network centrality of each molecule belonging to each cluster we identied top 3 combined signatures We also highlighted cluster-2 (hsa-mir-944, CFTR, GABRB2, HNF4G, TAC1, and C7orf57) for its high cohesiveness and contained a combined signature. We then applied three well-known classiers (viz., SVM, KNN, and random forest) using 10-fold crossvalidation, and obtained a high AUC score for cluster-2. Finally, we conducted a survival study with each molecule of the same cluster. Our proposed combined signature detection strategy can determine the signature(s) for any microarray or RNA-Seq prole. The code is available at https:// github.com/sahasuparna/DeMoS Keywords: Co-expression, Limma, Combinatorial transcriptomic fused signature, Disease classication, miRNA target gene, survival study
Cite This Article
Saha S, Mallik S, Bandyopadhyay S. A Multi-Molecular Fusion to Detect Transcriptomic Signature in Tissue-Specific Cancer. EJMO. 2022; 6(2): 156-171
Corresponding Author: Saurav Mallik