Motivation: Kidney renal clear cell carcinoma, which is a common type and accounts for 70-80% of renal cell carcinoma, can easily lead to metastasis and even death. A reliable signature for diagnosis of this cancer is in need. Hence, we seek to select miRNAs for identifying kidney renal clear cell carcinoma. Method: A feature selection strategy is used and improved to identify microRNAs for diagnosis of kidney renal clear cell carcinoma. Samples representing kidney renal clear cell carcinoma and normal tissues are split into training and testing groups. Accumulated scores representing the variable importance of each miRNA are derived from an iteration of resampling, training, and scoring. Those miRNAs with higher scores are selected based on the Gaussian mixture model. The sample split is repeated ten times to get more central miRNAs. Results: A total of 611 samples are downloaded from TCGA, each of which contains 1,343 miRNAs. The improved feature selection method is implemented, and five miRNAs are identified as a biomarker for diagnosis of kidney renal clear cell carcinoma. GSE151419 and GSE151423 are selected as the independent testing sets. Experimental results indicate the effectiveness of the selected signature. Both data-driven measurements and knowledge-driven evidence are given to show the effectiveness of our selection results.

diagnosis; feature selection; kidney renal clear cell carcinoma; miRNA; signature

Harbin Institute of Technology