Host mRNA m6A methylation modifications filtered

[wangguiqian]

How are false positive methylation modifications filtered out in the final CHEUI output files m5A_model2_predictions.txt and m5C_model2_predictions.txt? I'm using the RNA002 module for Direct-RNA sequencing and the rna_r9.4.1_70bps_hac_prom.cfg configuration file to obtain FAST5 files. My data is mRNA obtained from pig cells infected with a large double-stranded DNA virus. How can I accurately filter out methylation modifications from this host?

My filtering criteria for Mode2 files were: Probability > 0.95; Stoichiometry > 0.5, but the final methylation modification conditions looked like this:

Therefore, I feel there's a problem. Normally, m6A methylation modification of mRNA in host cells is associated with the DRACH motif. Why doesn't it look like this after filtering? Is it a problem with my processing? Or what's going on?

Thank you for your help. I also used the m6ABasecaller software and found that the methylation modification site information was completely different. Could there be a problem with my analysis?
Thank you for your help.
Best all,

[EduEyras]

Dear Kevin

Thanks a lot for your message and for using CHEUI.

If you can, please use SWARM (https://github.com/comprna/SWARM)

We improved the m6A detection over CHEUI and all other methods for RNA002. Also SWARM is actively maintained.

In both CHEUI and SWARM we measure m6A probability at every A, also outside of DRACH sites. We did this to allow unbiased m6A discovery in any species. But it is true that in mammals m6A should be mostly in DRACH.

To reduce false positives you could just restrict the sites to those with the DRACH motif. SWARM controls FPs better, and most of the sites will be DRACH. But still, you could simple look only at DRACH sites. Most of them will be the expected GGACT, etc...

In a recent analysis of m6A conservation in mammals, we also observed conserved m6A sites in non-DRACH AG-rich sites, potentially controlled by METTL16. An idea could be to separate the sites you detect into DRACH and non-DRACH. DRACH sites would be explained by METLL3 activity (or downregulation of erasers, etc...). It could be interesting if there is a large group of on-DRACH sites that are similar to the METTL16 motif. An alternative of course is that these are actually m1A sites. We showed in the CHEUI paper that nanopore signals for m1A and m6A are very similar. m1A also has some described motifs, which could help you separate them from the others.

I hope this helps

Best

Eduardo

[wangguiqian]

Dear Eduardo

You mentioned detecting non-DRACH sites. Yesterday, I reset the double restriction to 0.1 and 0.9, and upon running the test again, I found that a large number of new non-DRACH sites were generated. I personally think this might be what you meant. Also, my data comes from mRNA obtained from host pig cells infected with a large DNA virus, and I performed RNA methylation modification detection. I found a problem: even with different restriction settings, I still couldn't accurately find the active methylation modification, but I found that only 25% of the output data was in DRACH format. I have another question for you. My main goal is to study the methylation of mRNA produced by this virus-infected host cells. I want to verify the accuracy of the final output methylation modification sites by first detecting whether the host's mRNA methylation conforms to DRACH sites, which can effectively demonstrate the accuracy of mRNA methylation produced by viral transcription.
This is the output after I set the constraints. This is the statistical information on the methylation site motifs of host cell mRNA:

Best all
kevin

[EduEyras]

Hi Kevin,

CHEUI and now also SWARM (https://github.com/comprna/SWARM) will test all A's for m6A modification. DRACH is the most common m6A-modified motif in mRNA in mammals, modified by the METTL3 complex, but there are others, such as an AG-rich motif modified by METTL16. We found that there may be more METTL16-dependent sites than initially suspected. Also, there are variations of the DRACH motif, such as RRACH DRAC or RAC, which have been described to be METTL3 dependent as well.
There is also the potential confounding effect of m1A, as I mentioned. But it was described to be rare within mRNA.

I am not sure I understand what you mean by accuracy. The analysis depends on your specific hypothesis.

If you know for sure that METTL3 (or a complex member) is going up during infection, or that an eraser is going down, then you would expect DRACH sites to increase stoichiometry. But then your test would be to specifically test whether DRACH sites (modified by METTL3) have a significant increase of stoichiometry in the infection condition.

If you don't know whether METTL3 or any DRACH-related protein changes expression, you could do the same test. If DRACH sites increase methylation stoichiometry, then METTL3 might be involved.

But it could be that only AG-rich sites go up in stoichiometry, which would suggest that it is METTL16 the responsible one.

It could help to first identify all the A sites that are m6A with some level of confidence in either condition, and then perform a differential stoichiometry test on those sites (as long as they have read coverage in both conditions) to select sites that change significantly. Then you can investigate whether the changes occur predominantly in a speific motif type. You could e.g. run MEME or a similar motif finding tool.

I hope this helps

best

Eduardo

[wangguiqian]

Thank you for your reply. When I run the SWARM_read_level.py script in the SWARM package, do I need to set the --model1 parameter to add the model1 file? Or have you already specified the path to the original model1 file?

Below is the script command I ran:
python3 SWARM_read_level.py -m m6A --sam Forward_eventalign.sam

--fasta genome_HLJ.fa

--raw ${BLOW5_FILE}

--out ${OUT_FWD}

--kit RNA002

--threads ${THREADS}

[wangguiqian]

Thank you for your help. I have carefully reviewed your Python script and noticed that you have automatically added the original model1 and model2 files. However, I have a question: I am using the rna_r9.4.1_70bps_hac_prom.cfg sequencing instrument platform. Are the electrical signals from nanopore sequencing and Mini platforms different from those from my platform?
For direct RNA sequencing without a reference to detect methylation, could errors in the electrical signal due to the original consecutive identical base sequences significantly affect the judgment of methylation modifications?

[EduEyras]

Hi Kevin Thanks for the message Please post your comment/question in the SWARM github https://github.com/comprna/SWARM Thanks! Eduardo

…

On Wed, 25 Feb 2026 at 20:22, kevin Wang ***@***.***> wrote: *wangguiqian* left a comment (comprna/CHEUI#47) <#47 (comment)> Thank you for your help. I have carefully reviewed your Python script and noticed that you have automatically added the original model1 and model2 files. However, I have a question: I am using the rna_r9.4.1_70bps_hac_prom.cfg sequencing instrument platform. Are the electrical signals from nanopore sequencing and Mini platforms different from those from my platform? For direct RNA sequencing without a reference to detect methylation, could errors in the electrical signal due to the original consecutive identical base sequences significantly affect the judgment of methylation modifications? — Reply to this email directly, view it on GitHub <#47 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADCZKB77DBVWMLLACFQCZKT4NVSVLAVCNFSM6AAAAACUMVNHBOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTSNJXHA4TSNBSHE> . You are receiving this because you commented.Message ID: ***@***.***>

[wangguiqian]

comprna/SWARM#4

Hi Kevin

Thanks for the message
Please post your comment/question in the SWARM github
https://github.com/comprna/SWARM

Thanks!

Eduardo
…
Dear Eduardo,
I have carefully reviewed the principles behind your method for identifying methylation. Your judgment mainly involves examining the signal differences between the original 5 consecutive bases and the signal offset of the standard unmethylated data of the control 5mer. However, I have a question: Could the 5 bases with the original standard control signal offset be added to the standard 10-15 base length pattern to obtain the different electrical signal offsets corresponding to each site of this long sequence base? Would this optimize the judgment?

Best
Kevin

[wangguiqian]

Dear Kevin

Thanks a lot for your message and for using CHEUI.

If you can, please use SWARM (https://github.com/comprna/SWARM)

We improved the m6A detection over CHEUI and all other methods for RNA002. Also SWARM is actively maintained.

In both CHEUI and SWARM we measure m6A probability at every A, also outside of DRACH sites. We did this to allow unbiased m6A discovery in any species. But it is true that in mammals m6A should be mostly in DRACH.

To reduce false positives you could just restrict the sites to those with the DRACH motif. SWARM controls FPs better, and most of the sites will be DRACH. But still, you could simple look only at DRACH sites. Most of them will be the expected GGACT, etc...

In a recent analysis of m6A conservation in mammals, we also observed conserved m6A sites in non-DRACH AG-rich sites, potentially controlled by METTL16. An idea could be to separate the sites you detect into DRACH and non-DRACH. DRACH sites would be explained by METLL3 activity (or downregulation of erasers, etc...). It could be interesting if there is a large group of on-DRACH sites that are similar to the METTL16 motif. An alternative of course is that these are actually m1A sites. We showed in the CHEUI paper that nanopore signals for m1A and m6A are very similar. m1A also has some described motifs, which could help you separate them from the others.

I hope this helps

Best

Eduardo

Hi Eduardo,
Now that I've located the methylation site, how do I go back and find the original Fastq sequence ID of that methylation site? I want to statistically analyze the differences in PolyA tail length between methylated and unmethylated reads.

Best，
Kevin