A variational algorithm to detect the clonal copy number substructure of tumors from scRNA-seq data

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding tumor microenvironment composition and heterogeneity.
• Analyzing the clonal substructure of tumors from scRNA-seq data presents computational challenges.

Purpose of the Study:

• To introduce Single CEll Variational ANeuploidy analysis (SCEVAN), a novel variational algorithm for deconvoluting tumor clonal substructure from scRNA-seq data.
• To provide a practical framework for analyzing tumors and their microenvironment by accurately discriminating malignant and non-malignant cells.

Main Methods:

• SCEVAN employs a multichannel segmentation algorithm based on shared breakpoints across cells within the same copy number clone.
• It utilizes smoothed expression profiles of individual cells as evidence for subclone copy number profiles.
• The algorithm is designed for speed and accuracy in analyzing large scRNA-seq datasets.

Main Results:

• SCEVAN successfully deconvolves the clonal substructure of tumors from scRNA-seq data.
• The algorithm accurately and automatically discriminates between malignant and non-malignant cells.
• Application to 106 samples (93,322 cells) across diverse tumor types and technologies demonstrated its robustness.

Conclusions:

• SCEVAN offers a fast and accurate method for analyzing tumor clonal substructure and heterogeneity.
• It provides a practical tool for characterizing the tumor microenvironment and understanding geographic evolution in malignant brain tumors.