Question about funcitonal comparison between GEMs

[JOELROJAS2022]

I was using your functional comparison tutorial (https://sysbiochalmers.github.io/Human-GEM-guide/gem_comparison/) with my own context-specific GEMs generated with Human1. When using the "compareMultipleModels" function all of the models fail the 256 tasks from the full metabolic task document.

This resulted in 0 functional differences between the models.

I was able to run every other command prior to this, including the ftINIT extraction. The structure of the extracted model seems fine. What do you think might be causing this issue?

[JHL-452b]

Hi there! From the error messages ("Could not find all inputs in [...]"), it looks like the function can’t find the required inputs for the tasks listed in your full metabolic task document.

A couple of quick checks that might help:
First, double-check that the task definitions in your full metabolic task document are complete and match the metabolite IDs or Names used in your models. I haven’t seen this exact error before, so it’s possible there’s a mismatch or typo in the task definitions.

Then, before running any task-based function like this, it’s essential to run closeModel on your models (a function from RAVEN, you can use it like model = closeModel(model);). This function adds boundary metabolites and their exchange reactions, which are required for the tasks to work properly.

[JOELROJAS2022]

Thank you, you were right, it was due to a different symbol for the extracelular compartment. I was wondering if I am supposed to use normalized transcriptomic data for the model extraction process? In my case I used the raw TPM. Also, the tutorial for GEM extraction for Human1 included only ftINIT (https://sysbiochalmers.github.io/Human-GEM-guide/gem_comparison/), so I wanted to know if this is the recommended method for Human1, compared to iMAT, MBA or GIMME?

[JHL-452b]

Thanks for the questions. Using raw TPM values is generally acceptable for Human1 model extraction, and many Human-GEM workflows are based on TPM-like expression data. The tutorials mainly use ftINIT because Human1 was developed together with the INIT/tINIT/ftINIT framework, which is particularly suitable for large human metabolic models and tends to generate more functional, task-consistent tissue models.
iMAT, MBA, and GIMME are all widely used context-specific reconstruction methods, and their performance can vary depending on the biological question, the quality of the transcriptomic data, and how expression thresholds are defined. So, I think exploring different approaches and comparing the resulting models is often informative.

[edkerk]

Thank you, you were right, it was due to a different symbol for the extracelular compartment.

@JOELROJAS2022 Is this something that you manually implemented or did this come from a bug from RAVEN code? Just thinking if there is something that we should correct.