search

madagiurgiu25 / decoil-pre

Reconstruct ecDNA from long-read data using Decoil tool
BSD 3-Clause "New" or "Revised" License
7 stars 0 forks source link

ecDNA Filtering #11

Closed thomasjay3 closed 3 months ago

thomasjay3 commented 3 months ago

Hi,

I have a question about potential filtering steps for the decoil output.

I used Decoil on a sample and found 43 ecDNA, which was more than I was expecting. I have also used coral on the same long read sequencing dataset and found 3 ecDNA. amplicon architect was used on this sample with short read sequencing and found 3 ecDNAs as well.

I was curious if there are any parameters that you would recommend adjusting or potentially filtering steps that might be helpful.

I used the decoil-pipeline via Docker as outlined in the readme section. I haven't adjusted any of the optional arguments yet. I'm happy to provide additional information if that would be helpful.

Thank you so much!

madagiurgiu25 commented 3 months ago

Dear @thomasjay3,

Thank you for reaching you. For clarification, you considered only those calls which were labeled as "ecDNA" in the summary.txt? What is the WGS mean coverage of the sample?

The real data on which I have applied has ~5-10X mean coverage. How does the results look like if you downsample your sample to 5X WGS mean coverage?

thomasjay3 commented 3 months ago

Thanks for getting back to me! Yes, the WGS mean coverage of the sample is 27x. I will re-run with downsampling and get back to you.

thomasjay3 commented 3 months ago

I wanted to provide a quick update. Downsampling to 6.4x mean coverage reduced the predicted ecDNAs to 14. I think that 3 of these are true ecDNAs, and the others look like minor variations of these structures.

Also, the associated visualization tool is very and informative helped me better understand the structures in these samples. Thanks so much for this great tool and your assistance.

madagiurgiu25 commented 3 months ago

Dear @thomasjay3,

thank you for the updates. The default for --filter-score is set to 0 as of now, however setting it up to --filter-score 4xWGS would show you only structures which have proportion estimates larger than 4x times the WGS mean coverage. Additionally, if your sample is heavily rearranged, you might increase the --min-vaf to 0.1 (default is very low 0.01).

eesiribloom commented 1 month ago

when you say --filter-score 4xWGS do you mean run decoil with --filter-score 4 or if I have e.g. 30x mean coverage, I should run decoil with --filter-score 120 ?