Classify Cancers Types Using miRNAs Profiles and Machine Learning

Cancer classification is important for early diagnosis and effective treatment. This study proposes a hybrid framework that combines Fisher Discriminant analysis (FD) with Linear Support Vector Machine (LSVM) and Long Short-Term Memory (LSTM) networks. MicroRNAs (miRNAs) expression profiles for three types of cancer are used to evaluate the model. These are: lung- adenocarcinoma, ovary and pancreas cancers. The classification process is done using 5-fold cross valuation. Results show significant performance for LSTM compared to LSVM with and without features reduction. However, using FD analysis results in high accuracy reached 0.97and 0.96 for LSTM and LSVM respectively, using 50 miRNAs. The accuracy of all models improved with increased feature number sets. Another test has been done to find the relation between these types of cancers based on the classification process. Most pancreas cancer samples are classified as lung- adenocarcinoma cancer compared to ovary cancer in an unseen test. Overall, the results demonstrate the effects of LSTM -FD as a dependable tool for cancer diagnosis. Further, our model detects original cancer like the ovary and pancreas accurately.