Differential expression and celltype analysis [All Cell]¶

1. Differential gene expression (DGE) analysis identifies genes that show statistically significant differences in expression levels across distinct cell populations or conditions. This analysis helps in identifying which cell types are most affected by a condition of interest such as a disease, and characterizing their functional signatures.

2. Differential Compositional analysis identifies Quantifies changes in the relative abundances of each cell type across conditions (e.g., case vs. control, time points, treatment groups). Reveals expansions or contractions of specific populations that may not be captured by gene-level analyses alone.

Here, we introduced omicverse.single.DEG and omicverse.single.DCT to performed these two analysis.

Data Preprocess¶

The data we use in the following example comes from [Haber et al., 2017]. It contains samples from the small intestinal epithelium of mice with different conditions. We first load the raw cell-level data. The dataset contains gene expressions of 9842 cells. They are annotated with their sample identifier (batch), the condition of the subjects and the type of each cell (cell_label).

For our first example, we want to look at how the Salmonella infection influences the cell composition. Therefore, we create a subset of our compositional data that only contains the healthy and Salmonella-infected samples as a new data modality.

DEG with wilcoxon/t-test¶

In omicverse, we only need one function ov.single.DEG to perfrom all DEG analysis tasks. First, I will introduce the nonparametric Wilcoxon test and the t-test—two widely used methods for differential expression analysis. Due to their high computational efficiency, we applied them in our DEG analysis.

We need to set the ctrl_group and test_group to perform the analysis. Besides, we also need to define the celltype to be explode. If you set celltype_group is None, all the celltype will be calculated.

We can use sc.pl.violin to compare the gene expression between different condition.

DEG with memento¶

memento is a python package for performing differential mean, variability, and correlation in single-cell RNA sequencing data.

Memento, an end-to-end method that implements a hierarchical model for estimating mean, residual variance, and gene correlation from scRNA-seq data and provides a statistical framework for hypothesis testing of these parameters.

DCT with scCODA¶

In omicverse, we only need one function ov.single.DCT to perfrom all DCT analysis tasks. We included scCODA and milo to perform the celltype abundance analysis.

Besides, you can also perform the analysis with pertpy's api. dct_obj.model will be helpful.

More tutorial could be found in https://pertpy.readthedocs.io/en/stable/tutorials/notebooks/sccoda.html

No-U-turn HMC sampling is then initiated by calling sccoda_model.run_nuts().

We can use help(dct_obj.model.run_nuts) to obtain the argument as input.

Adjusting the False discovery rate¶

scCODA selects credible effects based on their inclusion probability. The cutoff between credible and non-credible effects depends on the desired false discovery rate (FDR). A smaller FDR value will produce more conservative results, but might miss some effects, while a larger FDR value selects more effects at the cost of a larger number of false discoveries.

The desired FDR level can be easily set after inference via sim_results.set_fdr(). Per default, the value is 0.05, but we recommend to increase it up to 0.2 if no effects are found at a more conservative level.

In our example, setting a desired FDR of 0.4 reveals small effects on Endocrine and Tuft cells. Keep in mind that we chose this value only for instructive purposes, since there are no credible effects beside Enterocytes at lower FDR levels. In practice, expecting 40% of all credible effects to be false-positives is usually not recommended.

DCT with milo¶

Many biological conditions (disease, development, genetic KOs) can induce shifts in cell composition, where cells of a given state become enriched or depleted in response to a perturbation. With differential abundance analysis, we quantify consistent changes in cell composition across replicate samples. While differential abundance analysis can be performed on cell type clusters, it’s not always possible or practical to use precisely labeled clusters, especially when we are interested in studying transitional states, such as during developmental processes, or when we expect only a subpopulation of a cell type to be affected by the condition of interest. Milo is a method to detect compositional changes occurring in smaller subpopulations of cells, defined as neighbourhoods over the k-nearest neighbor (KNN) graph of cell-cell similarities.

Tutorials could be found in https://pertpy.readthedocs.io/en/stable/tutorials/notebooks/milo.html

Build KNN graph¶

We can use omicverse functions to build a KNN graph. We set the dimensionality and value for k to use in subsequent steps.

Here the value of k indicates the smallest possible size of neighbourhood in which we will quantify differential abundance (i.e. with k=50 the smallest neighbourhood will have 50 cells). Depending on the number of samples, you might want to use a high value of k for neighbourhood analysis, to have sufficient power to estimate abundance fold-changes. Since here we have data from > 100 patients, we set k=150 to have on average more than one cell per donor in each neighbourhood.

Differential abundance testing with GLM¶

Similar to the scCODA approach, we use omicverse.single.DCT to perform differential cell–abundance analysis, except that here we set the method to milo. In future releases, we will gradually remove the rpy2‐based edgeR dependency so that milo analyses can run entirely in a native Python environment.

We can start inspecting the results of our DA analysis from a couple of standard diagnostic plots.

We can see that for the majority of neighbourhoods, almost all cells have the same cell type label. We can rename neighbourhoods where less than 60% of the cells have the top label as “Mixed”

Visualization¶

This is my favorite Milo plot. First, we create a color_dict to specify the cell colors we want to visualize.