Skip to content

Cell stratification and CNV#

Motivation#

Distinguishing malignant cells from normal cells remains a significant challenge in single-cell transcriptomics. To confidently identify malignant cells among other cell populations, multiple pieces of evidence are necessary. In our pipeline, we have incorporated two common types of evidence: gene signatures and CNV analysis.

Step-by-step#

In this module, we employ a conservative strategy, offering users the choice between two distinct methods for identifying malignant populations: i) the inferCNV-based method, and ii) the consensus score. The consensus score is a method currently under development, which amalgamates gene signatures, various CNV predictions, and cluster/patient specificity.

1. Running pipeline#

1.1. On the HPC#

By default the previous command line considers thresholds.

HPC

  • workflow_level = Stratification
  • input_stratification_method = infercnv_label
  • thr_cluster_size = 1000
  • thr_consensus_score = 2
nextflow run main.nf --workflow_level Stratification --project_name Training --sample_csv sample_table.csv --meta_data meta_data.csv --cancer_type Ovarian -resume -profile seadragon

1.2. On Cirro#

Alternatively, we execute this task on Cirro.

Cirro

  • Defining the pipeline entrypoints = Stratification
  • Method to define stratification labels = infercnv_label
  • Defining cluster size limit = 1000
  • Consensus score threshold (Beta) = 2

On Cirro, users should (Do not run):

  • Navigate to the Pipelines tab and enter "BTC scRNA Pipeline" in the search engine.
  • Change the Dataset to BTC Training dataset and the Copy Parameters From option to Run_01.
  • Double-check the aforementioned parameters and click Run.

2. Inspecting report#

For convenience the figures can be located in the Test_stratification_report.html report within the Run_02 dataset.

2.1. InferCNV predictions#

The first UMAP displays malignancy status based on inferCNV predictions. By default, the pipeline will utilize this annotation for downstream analysis. The inferCNV utilizes a time-consuming statistical method, therefore we applied a heuristic approach by subsampling cells per cluster, thr_cluster_size. In summary, the method performs well, but it has limitations in displaying misassigned cells, and low accuracy in small populations.

Image caption

Warning

InferCNV predictions are carried out in two distinct modes: reference-based, which leverages known malignant cells from the single-cell experiment, and reference-free mode, which employs a statistical procedure to estimate the CNV load cut-off. Currently, we utilize the reference-free mode. The pipeline stores the inferCNV heatmap in the data folder.

2.2. Consensus approach (under development)#

Alternatively, we also offer annotations based on the consensus approach (under development), which appears to more effectively capture smaller cell populations. However, neither approach can be considered flawless.

Image caption

Info

The user can change malignancy classifier by switching the parameter, input_stratification_method. The options are infercnv_label or consensus_label.

2.3. CD45 marker based on FACS#

Image caption

The malignancy prediction can generally be correlated with CD45 status (protein-level expression). However, minor discrepancies might be linked to the presence of normal cells with CD45- status in the dataset.

3. Exercise: Exploring alternative approaches to perform malignant identification#

Question

Does the consensus method affect the meta-program analysis? What happens if we change the consensus threshold? A: Run_Consensus and Run_Consensus_Meta_Threshold

Please note: When configuring the pipeline on Cirro, ensure that the Dataset is set to BTC Training dataset and select Run_02 for the Copy Parameters From option. Additionally, configure the Entrypoint parameter to Stratification.

Tip: Accelerate the process by reducing cluster size limit to 100

Reference#

  1. Infer Copy Number Variation from Single-Cell RNA-Seq Data