Flux Balance Analysis (Part 3)

Introduction

Multimodality of single-cell analysis

 

`Single-cell analysis` initially focused on sequencing mRNA at single-cell resolution. Advances in technology now allow us to quantify intracellular and cell surface proteins, DNA methylation, and chromatin accessibility at the cellular level.

 

However, single-cell metabolomics has stagnated. As mentioned in Part 2, quantifying metabolites is challenging due to their diversity, including sugars, nucleic acids, amino acids, fatty acids, and more. A human cell contains approximately 222,000 unique metabolites.

 

Figure from Rappez et al., (2021)

 

In 2021, single-cell metabolomics made a breakthrough by quantifying 740 metabolites from 30,000 cells. However, this number is still low compared to the approximately 222,000 unique metabolites in human cells, and as far as I know, no significant advances have been made since then.

 

 

To overcome the challenges of measuring metabolites at single-cell resolution, the research group developed a computational algorithm called `Compass`. Compass uses single-cell RNA sequencing data to predict each cell's metabolic state.

 

Using Compass on in vitro T helper 17 (Th17) cells, the group discovered a novel metabolic pathway that contributes to the autoimmune pathogenicity of Th17 cells.

 

Compass

The Compass algorithm uses linear programming to determine each cell's optimal reaction speeds.

 

To integrate gene expression data, we maximize the `weighted sum of reaction speeds`, where each weight $p$ is the toal mRNA involved in that reaction.

 

 

For example, consider two cells and three genes: Genes 1 and 2 participate in Reaction 1 and Gene 3 participates in Reaction 2.

 

By summing the mRNAs for each reaction, we calculate an evidence score for each reaction. Applying the inversion function, $ \frac{1}{1+R} $ creates a `penalty matrix`, assigning lower penalties to reactions with high evidence. Selecting the convex with the lowest weighted sum identifies the cellular state where gene expression most supports reaction speed.

 

Pseudocode of Compass

 

Note that this explanation focuses solely on the flux balance analysis component of Compass. Refer to the original article for detailed methodology.

 

Summary of results

 

After applying the Compass algorithm to scRNA-seq data, the researchers compared pathway enrichment analyses based on genes and reactions. The left graph illustrates traditional differentially expressed gene analysis, while the right graph displays differential reaction analysis from Compass.

 

The comparison shows that Compass provides more interpretable results, as most reaction speeds identified by Compass are concentrated in a single phenotype.

 

 

Additionally, fold changes in reaction speeds can be visualized in a metabolic network graph, making it easier to interpret differential metabolic pathway enrichment between groups.

 

One huge caveat of Compass is that it is best when applied to in vitro single-cell RNA sequencing data. Compass "computations assume an environment rich with nutrients, which accords with the studied in vitro growth media. A more accurate representation of the cellular environment should increase the algorithm’s predictive capabilities."