Quality Scores

[nadiaimam]

Hello,
I was wondering if there's any way to assess the quality or confidence of amplicons that are generated. I looked through the amplicon and classifier output folders but I didn't see anything that may help me gauge the confidence of the amplifications or ecDNAs detected.
Thanks!

[jluebeck]

Hi, this is a great query.

In the [prefix]_feature_basic_properties.tsv file, there is a column that flags low CN ecDNA, which can act as an initial filter.

If you want only the most confident ecDNA predicted by AC, you can also increase the stringency of the classification by setting the following:

--decomposition_strictness [0 to 1] # how strictly to filter low CN decompositions, default value is 0.1, but I suggest trying 0.25

--min_flow [value] # minimum cycle copy count to consider when identifying focal amps, default 1, but I suggest experimenting with values in the range of 1 to 4 for more stringent calls.

Another way to assess quality could be to consider how much of the genome is placed into ecDNA bed files for an AA amplicon, versus placed into Unknown bed files for the same amplicon.

We are also happy to help interpret outputs if you are interested.

Thanks,
Jens

[nadiaimam]

Thank you for your response! I would be interested in getting your opinion on why a certain sample comes up as BFB positive, while a similar sample comes up ecDNA positive. I used WGS data from a HeLa sample treated with 240 nM methotrexate and another HeLa sample treated with 600 nM methotrexate as what were assumed to be positive controls for ecDNA. After AA, I found that the 240 nM sample was indeed positive for ecDNA, but the 600 nM sample was only positive for BFB. I was surprised since the amplifications are in almost the same spot and the only change in condition was an increase in methotrexate concentration. Do you know why the higher concentration was called as BFB instead of ecDNA? I've included screenshots of the annotated cycles files incase that explains anything.

[jluebeck]

Hi, great questions again. Focal amplifications that are borderline for meeting the requirements of a classification often shift bioinformatic labels between replicates due to subtle changes in discovered SVs, estimated copy numbers, etc. Typically, BFBs with complex rearrangements can look more like ecDNA than BFBs, leading to misclassification. Further - the occurrence of a BFB may actually precede the formation of an ecDNA on that same location given the genomic instability inherent to BFB cycles, leading to cases where we see ecDNA that also have BFB-like properties. Sadly there is no easy bioinformatic fix for these problems using short reads.

The AA sashimi plots can also help with interpretation, and I would be happy to take a look at those for this case. Feel free to either share here or reach out via email: jluebeck@ucsd.edu.

Generally speaking, metaphase FISH staining the gene of interest and the native chromosome's centromere is a gold standard for resolving the final status of ecDNA.

Thanks,
Jens