Guidance on choosing the optimal threshold from RMT_threshold results

[Charlie-Hong-smu]

Hello,

First, I'd like to say thank you for creating the MetaNet package. It's a very powerful and useful tool for microbial network analysis.

I am currently using the RMT_threshold() function to determine an objective correlation threshold for my network, and I have a question about interpreting the results.

I've noticed that after running the function and plotting the result object, there appear to be three different potential thresholds:

1. The value stored directly in the returned list object: result$rmt_threshold.

2. The value printed to the console when using the generic plot() function: recommend r_threshold.
   For example, in one of my analyses, I observed the following:

3. The value stored in the object was ~0.22 (rmt_res_msi_h$r_threshold).

4. The plot() function printed a recommend r_threshold of ~0.24.
   This leads to my question: Could you please provide some guidance on which of these thresholds is generally recommended for constructing the final network for downstream analysis?

I would also be very grateful if you could briefly explain the difference in the calculation or logic behind the value stored in the object versus the one recommended by the plot() function. Understanding this would be very helpful for correctly interpreting the results and for writing the methods section of a publication.

Thank you again for your time and for this fantastic package.

[Asa12138]

Thank you for using this package.

Unfortunately, while working on this function, I consulted numerous RMT theory resources and found that they cannot directly provide a single definitive value, as there are multiple criteria for determining recommended values. Therefore, I suggest you consistently use one set of thresholds in your analysis and clearly state the criteria you adopted in your publication materials. For example: bigger log_LE, less log_LW, less log_see, bigger p_ks_test.

For the principles of threshold selection using RMT, you may refer to:
"Molecular ecological network analyses"
https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-13-113

For using the RMThreshold package, please see:
"RMThreshold"
http://matstat.org/content_en/RMT/RMThreshold_Intro.pdf

[Charlie-Hong-smu]

Thank you for using this package.

Unfortunately, while working on this function, I consulted numerous RMT theory resources and found that they cannot directly provide a single definitive value, as there are multiple criteria for determining recommended values. Therefore, I suggest you consistently use one set of thresholds in your analysis and clearly state the criteria you adopted in your publication materials. For example: bigger log_LE, less log_LW, less log_see, bigger p_ks_test.

For the principles of threshold selection using RMT, you may refer to: "Molecular ecological network analyses" https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-13-113

For using the RMThreshold package, please see: "RMThreshold" http://matstat.org/content_en/RMT/RMThreshold_Intro.pdf

Thanks a lot!!! You solved a big problem of mine!!!

[Charlie-Hong-smu]

Thank you for using this package.

Unfortunately, while working on this function, I consulted numerous RMT theory resources and found that they cannot directly provide a single definitive value, as there are multiple criteria for determining recommended values. Therefore, I suggest you consistently use one set of thresholds in your analysis and clearly state the criteria you adopted in your publication materials. For example: bigger log_LE, less log_LW, less log_see, bigger p_ks_test.

For the principles of threshold selection using RMT, you may refer to: "Molecular ecological network analyses" https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-13-113

For using the RMThreshold package, please see: "RMThreshold" http://matstat.org/content_en/RMT/RMThreshold_Intro.pdf

Sorry for bothering you again! But I have one more problem now

If I divide a dataset into two groups(R and NR), and I want to create a network for each of them, now should I calculate r threshold separately based on RMT theory for R and NR groups, and choose the r threshold, for example, match to the bigger log_LE? If I do so, then the r threshold would be different, maybe for R is 0.75 and for NR is 0.78. I wonder that if it is reasonable for the downstream analysis such as comparing some network properties between them? Or should I just calculate one r threshold only in the original dataset, and use the threshold in the groups divided from the original one?

Thank you again for your valuable time!

[Asa12138]

@Charlie-Hong-smu

Hello, you can refer to many articles that use MENAP (with RMT method). They do select different thresholds for different groups. For example, “Co-occurrence networks reveal more complexity than community composition in resistance and resilience of microbial communities”

[Charlie-Hong-smu]

I understand now!!
Thank you very much!

[Charlie-Hong-smu]

Hi,

First, thank you for developing this very useful package.

I am encountering an error when using the RMT_threshold() function on a Spearman correlation matrix derived from a microbiome dataset. The function appears to fail during its internal plotting process.

The issue:

The function runs through several thresholds and then stops with the following error:

Error in plot.window(xlim, ylim, "", ...) : 'ylim' values cannot be infinite
This suggests that for a certain threshold, the resulting metrics produce an infinite value (Inf), which the internal plotting function cannot handle.

What I have tried:

Based on standard R practices, I have tried to suppress the plotting to get only the calculation results, but both attempts failed:

Adding plot = FALSE to the function call resulted in an unused argument (plot = F) error.

Wrapping the function call in pdf(NULL) and dev.off() still resulted in the same 'ylim' values cannot be infinite error.

Question:

Is there a recommended way to bypass the internal plotting of RMT_threshold()? My primary goal is to obtain the calculation results (specifically, the res table to find the best threshold) and the plotting is not essential for my workflow.

Any advice would be greatly appreciated.

Thank you for your time.

[Asa12138]

@Charlie-Hong-smu
please update this package

devtools::install_github("Asa12138/MetaNet")

and use RMT_threshold(corr,out_dir = "./", plot=FALSE) to get the result data.frame

[Charlie-Hong-smu]

Thanks! I'll try.