Normalization Strategy for MRAS calculation

[ErickMUO]

Hi Metaflux Developers,

I've had the opportunity to explore your tool, and I must say it's quite impressive for investigating metabolic hypotheses. It's been running smoothly for me so far. However, as I delved into its functionalities, I found myself experimenting with various data preprocessing strategies. As expected, these strategies led to different outcomes.

Now, I'd like to seek your expert advice on which approach would be the most appropriate for MRAS (Metabolic Reaction Activity Scores) calculations. Here are the alternatives I've been exploring:

1. TPM (Transcripts Per Million): Using TPM values as input for MRAS calculations.

2. log2(TPM): Applying a logarithmic transformation to TPM values before performing MRAS calculations.

3. VST (Variance Stabilizing Transformation) Transformed Counts: Using VST-transformed counts as input for MRAS.

4. log2(TPM) with Lowly Expressed Genes Removed: Applying a logarithmic transformation to TPM values after removing genes with low expression levels.

Your insights on which of these alternatives aligns best with MRAS calculations within Metaflux would be greatly appreciated. I'm aware that each approach has its unique strengths and limitations, and I want to ensure I'm using the most appropriate method.

[kunstner]

Hi, I just used VST values because I mapped my data using STAR aligner. It seems the results are decent, but it needs some more evaluation. Would be nice, if the developer could chime in with some recommendation. Maybe, the normalization method does not matter at all. Best, Axel

[AMAwatermelon]

and in sc-data, would default normalization method in seurat be ok?

[SebastianMHJohn]

Hi, if you read the source code you will realize that they create a new Seuratobject and just apply default normalization.

Best wishes, Sebastian

[kennyyeo13]

Hi @kunstner ,

How did you run vst data? Would vst data produce some negative values?

Thanks Kenny