Survivability Prediction from Multiomics Cancerous Data with 1DCNN-LSTM-Attention Mechanism
Ambika Hazarika, Ansuman Kumar, Anindya Halde · 2024
The possibilities for survivability prediction to enhance patient treatment and outcomes is gaining attention in the field of health care. A human being's biological composition can be fully understood by utilizing multiomics data, which combines various biological information from different omics viz., genomics, transcriptomics, proteomics, and metabolomics. Considering the increasing popularity of Deep Learning (DL) algorithms as a means for recognizing significant patterns in complex and multi-dimensional information, in this study, we propose a novel method for predicting survivability using multiomics data by utilizing one dimensional convolutional neural network(1DCNN),Long Short-Term Memory (LSTM) networks and attention mechanism. Preprocessing the multiomics data using 1DCNN to extract relevant features and integration of the omics data is the first step in our approach. Then, in order to extract the complex patterns and temporal relationships present in multiomics data, we use LSTM networks. Moreover, in order to improve model understanding and concentrate on informative characteristics, we include an attention mechanism into LSTM output. Experimental results of the proposed method when compared to LSTM and other machine learning classifiers utilized here, indicate that LSTM with attention mechanism performs better in terms of accuracy and other performance evaluation measure (precision, recall and F1-score)on three different types of cancerous multiomics data viz., Breast, Kidney, and Lung.