Prediction of the Thromboembolic Syndrome: an Application of Artificial Neural Networks in Gene Expression Data Analysis

The aim of this study was to propose a method for improving the power of recognition and classification of thromboembolic syndrome based on the analysis of ‎ gene expression data using artificial neural networks. The studied method was performed on a dataset which contained data about 117 patients admitted to a hospital in Durham in 2009. Of all the studied patients, 66 patients were suffering from thromboembolic syndrome and 51 people were enrolled in the study as the control group. The gene expression level of 22277 was measured for all the samples and was entered into the model as the main variable. Due to the high number of variables, principal components analysis and auto-encoder neural network methods were used in order to reduce the dimension of data. The results showed that when using auto-encoder networks, the classification accuracy was 93.12. When using the PCA method to reduce the size of the data, the obtained accuracy was 78.26, and hence a significant difference in the accuracy of classification was observed. If auto-encoder network method is used, the sensitivity and specificity will be 92.58 and 93.68 and when PCA method is used, they will be 0.77 and 0.78 respectively. The results suggested that auto-encoder networks, compared with the PCA method, had a higher level of accuracy for the classification of thromboembolic syndrome status.