Özet:

Abstract Genetic information is necessary for studying the biological processes that, when interrupted, cause certain cancers to form. Although advances in sequencing technology have made it possible to record the nuances of gene interaction in several data formats, using these methods to identify, diagnose, and treat cancer remains difficult. Machine learning has helped researchers in a number of areas, including supervised and unsupervised learning, as well as gene identification, but the results have been less than visually satisfying. Using RNA-SEQ data from The Cancer Genome Atlas, this research focuses on multi-class classification of cancer, extraction of key characteristics, and identification of relevant genes for 10 different types of cancer.Tests conducted with the restricted hardware resources at hand have shown that these limitations do not always exclude the possibility of positive results. Stacked de-noising auto encoders were employed for feature extraction and biomarker identification, while 1D convolutional neural networks were used for classification. Both the recovered features and the relevant genes were used in the classification process, with the former typically performing better (about 94% accurate) than the latter (95% accurate). By using stacked denoising auto-encoders to construct matrix weights and features, we were able to identify common cancer-related pathways and their associated genes.

Anahtar Kelimeler: