Cancer Types Prediction Using Recurrent Type Networks

Large-scale cancer genomics data often comes in multiplatform and heterogeneous forms. These datasets impose great challenges in terms of the bioinformatics approach and computational algorithms. Numerous researchers have proposed to utilize this data to overcome several challenges, using classical machine learning algorithms as either the primary subject or a supporting element for cancer diagnosis and prognosis.

In this chapter, we will use some deep learning architectures for cancer type classification from a very-high-dimensional dataset curated from The Cancer Genome Atlas (TCGA). First, we will describe the dataset and perform some preprocessing such that the dataset can be fed to our networks. We will then see how to prepare our programming environment, before moving on to coding with an open source, deep learning library called Deeplearning4j (DL4J). First, we will revisit the Titanic survival prediction problem again using a Multilayer Perceptron (MLP) implementation from DL4J.

Then we will use an improved architecture of Recurrent Neural Networks (RNN) called Long Short-Term Memory (LSTM) for cancer type prediction. Finally, we will see some frequent questions related to this project and DL4J hyperparameters/nets tuning.

In a nutshell, we will be learning the following topics in the chapter:

• Deep learning in cancer genomics
• Cancer genomics dataset description
• Getting started with Deeplearning4j
• Developing a cancer type predictive model using LSTM-RNN
• Frequently asked questions