Different outputs of RTSI scores and metabolite abundances using the same input file

Hello!! I am running melonnpan predict using the melonnpan.test.data.txt file provided in the tutorial. However, I am getting different outputs of RTSI and relative abundances of the metabolites when I repeat the process using the same input file.
What might be the reason for this?
I have attached the RTSI scores obtained from different melonnpan predict runs using the same input file.
2nd output.txt (1.3 KB)
1st output.txt (1.3 KB)

Hi @paramartha_banerjee - this has now been resolved. Can you try again after re-installing the latest version of the software from GitHub?

Many thanks,

Thanks!! @himel.mallick I am now getting the same output when I repeat the melonnpan workflow using the melonnpan.test.data.txt file provided in the tutorial.
However, it is different from the output provided in the tutorial.
RTSI scores provided in tutorial

I have attached a screenshot of the RTSI scores provided in the tutorial and the one obtained by me.

Hi @paramartha_banerjee - great catch. The tutorial is based on the previous version of the software. I have just updated it to reflect the current version of the software.

Also, since I did not explain the changes in this version before, I am adding a bit more detail for any future references on this topic.

  • The original version of the software back in 2019 used rank-based inverse normal (RIN) transformation of the metagenomic gene data and used the R package GenABEL to perform this operation.

  • In the subsequent versions, GenABEL was removed from CRAN but we forced users to install an older version of the software using devtools.

  • Later in 2022, we found a different implementation of the RIN transformation as a standalone function but it used a different mechanism to break the ties than the original GenABEL function.

  • In particular, the RIN transformation must decide on how to handle ties when there are zeroes in the data which is almost always the case with metagenomic gene abundances.

  • We initially randomly assigned unique ranks which explains the run-to-run variations in the RTSI scores but in the latest version I turned the random assignment off to alleviate this issue.

In summary, with the current RIN transformation, the RTSI scores will still be a bit different than the original MelonnPan algorithm but it should overall be very similar with all other MelonnPan steps remaining the same.

I hope this clarifies the discrepancy between different versions,

Thank you so much!! @himel.mallick