Cell types annotation of cycling T lymphocytes in the tumor microenvironment

Students:

Kristina Zheltova, Anastasiia Mikhailichenko, Vsevolod Zvezdin*

​

Supervisor:

Sergey Isaev (Medical University of Vienna)

Main goal: Develop a method for determining types of dividing cells and its application for the analysis of T cells from the microenvironment of tumors of various origins in order to determine the cell type in which clonal expansion occurs.

Tasks:

• Collect and analyze of a large sample of scRNA-Seq datasets of the tumor microenvironment for further subtyping of dividing cells in them
• Develop and test three methods for classifying cycling T cells, determine the most suitable one

Introduction

Single-cell transcriptomics of T cells infiltrating tumors could be used to predict tumor progression, to prescribe appropriate therapy, and to study the disease in fundamental research.

It is especially important to know which groups of lymphocytes are cycling. However, the signal of cycling stage markers in such cells is stronger than the signal of cell type markers. That makes it hard to tell which cell type the cycling cell belongs to.

Datasets preparation for training and validation

We created a pypeline to process scRNA-Seq datasets for training and validation

GSE154826

35 - patients

77 - samples

126910 - CD3+ cells

11838 - TCR

GSE162500

11 - patients

17 - samples

54971 - CD3+ cells

27493 - TCR

Datasets preparation for training and validation

We collected and analyzed large sample of scRNA-Seq datasets of the tumor microenvironment for further subtyping of dividing cells in them

Three methods for classifying cycling T cells

Drop Genes

​

Remove the genes associated with the cell cycle

​

Perform classification using the k nearest neighbors method

​

​

Cell Cycle Scoring

​

Find the genes whose expression is most associated with the cell cycle

​

Perform classification using the k nearest neighbors method

​

Label Transfer

​

Using annotated dataset integrate the data of unknown cells with this existing reference dataset

​

Predict cell type based on the environment of unknown cells

• Algorithm based on label transfer gives the best predictions and improved the quality of predictions compared to the baseline

Further plans:

• Algorithm modification for more stable results
• Testing algorithms on more datasets
• Generalization of annotation results and biological interpretation

Results

Baseline method

​

k nearest neighbors classification on the dataset without any preprocessing of the cell cycle signal