Computational analysis of gene expression data using bidirectional long short-term memory for disease diagnosis

Classification of cancer samples from microarray gene expression data is one of the important areas of research today. Traditional supervised techniques may not produce desired classification results because of their inability of extracting appropriate feature(s). In this context, here in this article, classification of microarray data sets is done using bidirectional long short-term memory (Bi-LSTM), a type of recurrent deep neural network. The curse of dimensionality issue is also handled using correlation-based feature selection. Experiments were carried out on eight benchmark cancer microarray data sets. The performance of the proposed method in terms of different validity measures is compared with three other deep learning-based classifiers and four conventional classifiers. It is observed that Bi-LSTM outperformed all other classifiers for all data sets. Further, it is observed that out of the total 56 paired t tests carried out for the proposed Bi-LSTM method, 42 cases are found to be statistically significant.