Deep Learning and identification of Cancer Related Sub-networks

Deep Learning and identification of Cancer Related Sub-networks

Smulders, S.
Ottervanger, B.
Bakker, P.

Allahyar, A. (mentor)

Electrical Engineering, Mathematics and Computer Science

Intelligent Systems

Pattern Recognition and Bioinformatics

TI3806

2016-06-29

The breast cancer survival rate has improved significantly between 1975 and 2003. The primary improvement in treatment is due to subtyping, where the high complexity cancers are divided in types of cancer and subgroups. In a study by Dr. Parker it was found that by looking at mRNA only, the intrinsic subtypes of breast cancer could be identified. In our work we will create images out of gene expressions, which we use to train a convolutional neural network to classify the created images on cancer subgroup. This report describes an analysis of using convolutional neural networks on images created using the correlations between gene expressions. We focused on using deep learning to identify cancer related gene-gene interaction sub-networks. Prior research has shown that using gene expression assays can be used to improve classification of subtypes of cancer. We created two dimensional images, preserving the correlations between genes as distances in the image and used convolutional neural networks to classify the sub-types. These networks can do object recognition, and in the case of our images the object could be build up of sub-networks of genes.

http://resolver.tudelft.nl/uuid:3aa750b1-5a3c-4350-92d3-0555ed1c97c4

Student theses

bachelor thesis

(c) 2016 The Authors