Quantifying Intratumor Heterogeneity by Key Genes Selected Using Concrete Autoencoder

Abstract

The tumor cell population in cancer tissue has distinct molecular characteristics and exhibits different phenotypes, thus, resulting in different subpopulations. This phenomenon is known as Intratumor Heterogeneity (ITH), a major contributor to drug resistance, poor prognosis, etc. Therefore, quantifying the levels of ITH in cancer patients is essential, and many algorithms do so in different ways, using different types of omics data. DEPTH2 algorithm utilizes transcriptomic data to assess ITH scores and exhibits promising performance. However, it quantifies ITH using all genes, limiting the identification of ITH-related prognostic genes. We hypothesize that a subset of key genes is sufficient to quantify the ITH level, and this subset of key genes could be ITH-related prognostic genes. To prove our hypothesis, we propose an unsupervised deep learning-based framework using Concrete Autoencoder (CAE) to select a subset of cancer-specific key genes for ITH evaluation. For the experiment, we used gene expression profile data of breast, kidney, and lung cancer tumor cohorts from the TCGA repository. Multi-run CAE identified three sets of key genes for each cancer cohort. Comparing ITH scores derived from all genes and CAE-selected key genes showed similar prognostic outcomes. Subtypes of lung cancer displayed consistent ITH distributions for both gene sets. Based on these observations, it can be concluded that a subset of key genes, instead of all, is sufficient for ITH quantification. Our results also showed that many key genes are prognostically significant and can be used as therapeutic targets.