Tracking intratumoral heterogeneity in glioblastoma via regularized classification of single-cell RNA-Seq data

Abstract

Background: Understanding cellular and molecular heterogeneity in glioblastoma (GBM), the most common and aggressive primary brain malignancy, is a crucial step towards the development of effective therapies. Besides the inter-patient variability, the presence of multiple cell populations within tumors calls for the need to develop modeling strategies able to extract the molecular signatures driving tumor evolution and treatment failure. With the advances in single-cell RNA Sequencing (scRNA-Seq), tumors can now be dissected at the cell level, unveiling information from their life history to their clinical implications. Results: We propose a classification setting based on GBM scRNA-Seq data, through sparse logistic regression, where different cell populations (neoplastic and normal cells) are taken as classes. The goal is to identify gene features discriminating between the classes, but also those shared by different neoplastic clones. The latter will be approached via the network-based twiner regularizer to identify gene signatures shared by neoplastic cells from the tumor core and infiltrating neoplastic cells originated from the tumor periphery, as putative disease biomarkers to target multiple neoplastic clones. Our analysis is supported by the literature through the identification of several known molecular players in GBM. Moreover, the relevance of the selected genes was confirmed by their significance in the survival outcomes in bulk GBM RNA-Seq data, as well as their association with several Gene Ontology (GO) biological process terms. Conclusions:We presented a methodology intended to identify genes discriminating between GBM clones, but also those playing a similar role in different GBM neoplastic clones (including migrating cells), therefore potential targets for therapy research. Our results contribute to a deeper understanding on the genetic features behind GBM, by disclosing novel therapeutic directions accounting for GBM heterogeneity.

Discriminative feature of cells characterizes cell populations of interest by a small subset of genes

“…Among them, LASSO–logistic regression is a method that can construct an interpretable linear model and perform variable selection in one step. However, the gene clusters obtained by these discriminative methods have been mostly used as gene signatures in cell-type classification (e.g., a cell is normal or malignant 29 ), and no attempt has been made to interpret these gene signatures themselves biologically.…”

TCox: Correlation-Based Regularization Applied to Colorectal Cancer Survival Data

“…The is based on sparse logistic regression and enables the selection of gene signatures shared by two diseases in breast and prostate cancer. The correlation structure was also relevant to identify heterogeneity factors in glioblastoma [ 18 ]. Instead of trying to retrieve similar correlation patterns, promotes genes that exhibit distinct relationships between two groups, thus highlighting potential differences in the corresponding sub-networks.…”

“…Penalty terms based on centrality measures of the nodes (genes) in the network have been suggested, such as the degree, therefore penalizing the variables based on their role in the overall network [ 12 , 16 ], and also by promoting the smoothness of the parameters across adjacent nodes in the network [ 17 ]. Network-based regularizers built on the correlation between the variables in different groups have also been proposed [ 13 , 18 ]. The central premise is that biomolecular networks in different cancer or cell types exhibit distinct network-based correlation patterns that might be regarded as biomarkers for disease/cell typing, but also similarities whose relevance might be investigated in the definition of common therapies for distinct disease conditions.…”

The Role of Network Science in Glioblastoma

“…Under the regularized-based learning framework, a variant of the above correlation-based regularizers, the twin networks recovery ( ) penalty was introduced by Lopes et al. (2020) [ 123 , 124 ] to explore the differences in the gene correlation networks in a classification setting. Given two disease types, the goal is to select features that have a similar correlation pattern across the two conditions, which can be regarded as putative disease targets for the development of shared therapeutic approaches for the two diseases.…”

“…The suitability of the above network-based regularizers has been shown to be promising when tackling GBM heterogeneity via biomarker selection and classification [ 123 ] in a single-cell RNA sequencing (scRNA-seq) GBM dataset [ 126 ]. The authors proposed a classification setting through sparse logistic regression to classify cells into different populations (neoplastic and normal cells), while selecting gene features discriminating between classes, but also those shared by different neoplastic clones (tumor core and infiltrating cells), standing as putative therapeutic markers to target multiple clones.…”