A Deep Learning Based Metastasis Prediction System Using Multi Omics Data and Recursive Neural Network

Nowadays, metastasis is the leading cause of tumor related deaths, as it involves a complex sequence of cellular events. Where the tumor cells break away from the primary site, enter the bloodstream, and are transported to a secondary site, where they invade and establish new growth. However, the existing heterogeneous ensemble model for breast cancer prediction, but it has required various algorithms for accurate prediction. Hence, this research proposes Recursive Neural Network (RecNN) to increase the accuracy of multi-omics data and effective metastasis prediction. Initially, the data is gathered from Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) dataset, a complete source of breast cancer genomic data. Then, the data is further preprocessed by thorough cleaning and Named Entity Recognition (NER) to categorize the named entities in unstructured text into predefined categories. After that, the Least Absolute Shrinkage and Selection Operator (LASSO) is utilized to select optimal features from the given data. Finally, RecNN is introduced to predict the metastasis according to the selected features. From the results, the proposed RecNN approach provided outperforming results in the accuracy is 98.69%, F1-score is 99.78%, precision is 98.64%, recall is 99.67% when compared with existing Convolutional Neural Network (CNN) model.