Prediction of Child Tumours from Microarray Gene Expression Data Through Parallel Gene Selection and Classification on Spark

Abstract

Microarray gene expression data play a major role in predicting chronic disease at an early stage. It also helps to identify the most appropriate drug for curing the disease. Such microarray gene expression data is huge in volume to handle. All gene expressions are not necessary to predict a disease. Gene selection approaches pick only genes that play a prominent role in detecting a disease and drug for the same. In order to handle huge gene expression data, gene selection algorithms can be executed in parallel programming frameworks such as Hadoop Mapreduce and Spark. Paediatric cancer is a threatening illness that affects children at age of 0–14 years. It is very much necessary to identify child tumours at early stage to save the lives of children. So the authors investigate on paediatric cancer gene data to identify the optimal genes that cause cancer in children. The authors propose to execute parallel Chi-Square gene selection algorithm on Spark, selected genes are evaluated using parallel logistic regression and support vector machine (SVM) for Binary classification on Spark Machine Learning library (Spark MLlib) and compare the accuracy of prediction and classification respectively. The results show that parallel Chi-Square selection followed by parallel logistic regression and SVM provide better accuracy compared to accuracy obtained with complete set of gene expression data.