CNVs are not detected about numbat HOT 4 CLOSED

Hello,

Thanks for trying out Numbat. Could you attach the bulk_subtrees_1.png as well?

In general, there are three scenarios that a CNV is not called. I'm thinking of making a FAQ about this since many people have asked.

1. The CNV is not called in pseudobulk HMM. You can check bulk_subtrees_*.png to see if the event was called. If it was not, I would suggest trying to find a better expression reference to reduce the noise in logFC, or changing the parameters used to configure the HMM (gamma, t, etc). Sometimes the CNV is very subclonal and was not isolated as a pseudobulk by the initial clustering. You can see if this was the case by looking at the exp_roll_clust.png and if so, try increasing k_init.

2. CNV is called in pseudobulks but did not appear in segs_consensus_*.tsv and joint_post_*tsv. This is because the event is filtered out due to low LLR. Lowering min_LLR threshold will help.

3. CNV is called in pseudobulks, is present in segs_consensus_*.tsv and joint_post_*tsv, but did not appear in phylogeny. This is because the event is filtered out due to high entropy (weak evidence) in single cells. Raising max_entropy threshold will help. You can check the entropy of specific events in joint_post_*tsv in the column avg_entropy.

Best,
Teng

from numbat.

Thanks @teng-gao, for the detailed answer! Below the bulk_subtrees_1.png

I will give a try to the several recommendations/options you have given!

It is a wonderful tool, and I am really happy with the performance! And kudos for the publication!

from numbat.

Thanks @ccruizm! Based on the pseudobulk profiles, the allelic frequency is quite balanced on all chromosomes so I suspect there aren't actually CNVs there (does WES show allelic imbalance?). The only other possibility is that all CNVs are balanced gains and losses, which is extremely rare.

from numbat.

I am waiting for the WES report. In the meantime, I checked the exp_roll_clust.png, which looks pretty similar to the one from copyKAT.

I increased the k_init (10) and now it detected some iCNV but actually they seems to be inverted of what the expression heatmaps show.

What do you think might be the reason of this incongruency? Thanks again for your help!

from numbat.