Abstract

Convolutional neural networks can learn hierarchical features directly from image pixels, making them powerful tools for image classification. This capability has motivated researchers to encode genomic and other non-image data as image-like representations that can be analyzed with image-based deep learning models. In particular, RNA-seq gene expression profiles have been transformed into two-dimensional images and classified by fine-tuning pretrained convolutional neural networks. Galaxy's Image Learner makes this general strategy accessible through a web interface in which users provide images, labels, and training settings.

About This Material

This is a Hands-on Tutorial from the GTN which is usable either for individual self-study, or as a teaching material in a classroom.

Questions this will address

• How can GTEx gene expression profiles be transformed into image-like inputs for Image Learner?
• How do GTEx sample annotations provide tissue labels for supervised classification?
• How can this workflow be run on any Galaxy server with Image Learner installed?

Learning Objectives

• Import a prepared GTEx v11 Image Learner metadata table and image ZIP archive.
• Optionally rebuild the prepared files from GTEx v11 gene TPM and sample annotation files.
• Train and evaluate a multi-class tissue classifier in Galaxy.