Firstly, thank you for the well documented methods on the wiki page!
I would like to know 2 things.
Where are the calibrated residual expressions in the output folder.
I have calculated the CNA score and correlation metrics using the residual expressions (re-centered at 0 instead of 1) . To assess how noisy the residual expression are - would it be more accurate to check if the peak overlaps with CNV level 1(normal) or compare how noisy the preliminary and final infercnv heatmap looks. Therefore, i can adjust the denoise parameter.
Below is the reason why i would like to look at the overlaps of peaks:
In one of the github issues (https://github.com/broadinstitute/infercnv/issues/228) one of the comments mentioned this-"If the values differ from 1, they will tend to show gain/loss in the directions you said yes. However, depending on the denoising settings/level you used, there can still be non-1 values that are just noise/tails of the base level expression distribution. You should look at the distribution of the values to determine whether there are peaks distinct from the base distribution or not." And directed the person to the HMM wiki link.
"For the different distributions, you can see the theory behind it in the wiki explanation of the HMM modelling https://github.com/broadinstitute/inferCNV/wiki/infercnv-i6-HMM-type#calibrating-cnvs-residual-intensity-distributions-via-the-hspike"
Dear Infercnv Team,
Firstly, thank you for the well documented methods on the wiki page!
I would like to know 2 things. Where are the calibrated residual expressions in the output folder. I have calculated the CNA score and correlation metrics using the residual expressions (re-centered at 0 instead of 1) . To assess how noisy the residual expression are - would it be more accurate to check if the peak overlaps with CNV level 1(normal) or compare how noisy the preliminary and final infercnv heatmap looks. Therefore, i can adjust the denoise parameter.
Below is the reason why i would like to look at the overlaps of peaks:
In one of the github issues (https://github.com/broadinstitute/infercnv/issues/228) one of the comments mentioned this-"If the values differ from 1, they will tend to show gain/loss in the directions you said yes. However, depending on the denoising settings/level you used, there can still be non-1 values that are just noise/tails of the base level expression distribution. You should look at the distribution of the values to determine whether there are peaks distinct from the base distribution or not." And directed the person to the HMM wiki link. "For the different distributions, you can see the theory behind it in the wiki explanation of the HMM modelling https://github.com/broadinstitute/inferCNV/wiki/infercnv-i6-HMM-type#calibrating-cnvs-residual-intensity-distributions-via-the-hspike"
Thank you! .