Deep Learning Methods to Classify Cancer vs. Normal Samples Using TCR Data

• Activity Number: 341 - Contributed Poster Presentations: Section on Statistical Learning and Data Science
• Type: Contributed
• Date/Time: Tuesday, August 9, 2022 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Yujia Cai* and Si Liu and Wei Sun
• Companies: Fred Hutchinson Cancer Research Center and Fred Hutchinson Cancer Research Center and Fred Hutchinson Cancer Research Center
• Keywords: Deep learning; Neural network; TCR sequencing; Confounding

Abstract

T-cell receptor (TCR) sequencing is emerging as a new type of profiling method to monitor human immune response to various stimuli, such as virus infection and tumor-associated antigens. We employed deep learning methods to classify three different types of samples: tumor samples and blood samples from either cancer or non-cancer patients, by leveraging the corresponding TCR-seq data of each sample. The biochemical features of amino acids were summarized and converted to construct an image of each TCR sequence, which was then fed into a neural network to classify samples. Our method has higher accuracy to classify tumor samples versus blood samples from cancer or non-cancer patients, but also has an encouraging performance to classify blood samples from cancer patients versus those from non-cancer patients. We will also discuss how to handle potential confounders when training our neural network.