K-mean clustering of miRNAs associated with cancer

• Published in: 2017 International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT) pp. 211 - 214
• Main Authors: Sankar, Janani, Thangavel, Dharani, Murugesan, Nivetha, Subramaniam, Nivedha, Kothandan, Ram
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2017
• Subjects: Biological system modeling; biomarker; Cancer; Clustering algorithms; Computational modeling; K-mean; miRNAs; Principal component analysis; Training; Tumors; Unsupervised learning
• DOI: 10.1109/ICICICT1.2017.8342561

The role of mirna in cancer has been an important development in tumour studies since its discovery in 2002. Recent studies, focuses much on small non-coding RNAs particularly miRNAs as biomarker for cancer detection and diagnosis; however the process is laborious. On the other hand, computational methods considers miRNA as a pivotal entity in cancer studies. However, considering the progenesis of cancer does not happen only with the miRNA alone. Several other physiological factors also favours the development of cancer in a cell. Here in this study, an attempt has been made to employ unsupervised learning algorithm - k-mean clustering algorithm to segregate miRNA as either oncogenic or tumour suppressor based on their interaction with the mRNA. Classification of miRNAs is mainly based on the sequence, thermodynamic and hybridization features extracted from miRNA-mRNA hybridized structures and miRNA sequences. Principal Component Analysis (PCA) was applied for feature processing. The distance between the clusters were computed using cosine similarity which was better than other distance measures. The performance of the model was evaluated using Davies-Bouldin index (DB index) which had a value of −1.5 to −2.0 which indicates the effectiveness of the model constructed.