Basic Informations

C.V

Master Title

A Proposed Intelligent Framework for Big Genome Data Analysis

Master Abstract

Women's cancers, which include breast adenocarcinoma and non-small-cell lung cancer, pose a serious threat to women's health. Female-oriented cancers are the leading cause of death among women in both developed and developing countries. The aggressiveness of these cancer types is caused by tumor infestation and an increased risk of metastasis. Women's cancers are emerging as the state of the art for molecular tumor classification, thanks to RNA-Seq data-based precision cancer diagnostic analysis, which is suitable for discovering new information on genes with differential expression. Since cancer-specific biomarkers can aid in the early identification and detection of cancer in women by using gene expression, such analysis is a tool for identifying fundamental patterns in data. This, in turn, aids in the development of treatment plans that will increase survival. The biggest hurdle in cancer detection is choosing the smallest number of informative genes. Highdimensional data from RNASeq experiments with small sample sizes is a problem. The learning model will therefore be over-fitted, which is the result of the dimensionality issue's repercussions and managing such a large dataset. In this thesis, we propose a new method for combining two different dimensionality reduction methods called FCBF-PCA: a filterbased approach for selecting informative genes using the FCBF algorithm and an unsupervised learning algorithm called principal component analysis (PCA) for feature extraction. We developed a different classification model that uses supervised learning with a deep learning algorithm called a feedforward neural network with hyper-parameters for optimization of the model. To identify the best subset of genes, the enhanced gene selector and feed-forward neural network classifier combine the statistical results of relevant genes with the calculation of gene dependencies using the symmetrical uncertainty (SU) measurement. And extensive packages to enable RNA-Seq analysis in R programming with Bioconductor. Six of its modules are: feature mapper, preprocessing gene expression, dimension transformer, exploratory data analysis, feature selector, deep learning technique, and machine predictors with hyper-parameters. Using biomarkers for prediction, with an accuracy of 96.7% for RNA-Seq breast adenocarcinoma data and 95.5% for RNA-Seq NSCLC cancer data, the deep learning algorithm is statistically significantly better in the metric when compared to other algorithms. Performance evaluation includes accuracy, sensitivity, specificity, precision, the F1 score, and the area under the curve (AUC) score. By utilizing deep learning, the suggested model can aid in the early identification and diagnosis of cancer in women and, consequently, assist in the design of initial treatment strategies to increase survival. The top 11 hub-gene biomarker discoveries that could be involved in non-small-cell lung and breast adenocarcinoma cancer are now presented. The results signify that our framework is statistically significantly better in the metric compared to other previous research studies.

PHD Title

Not registered yet

PHD Abstract

Not registered yet