ABsolute Immune Signal (ABIS) deconvolution output

Question

[harikrishnanc903]

I am currently working on the ABIS deconvolution method to find the immune cell type. I am not sure about the values obtained from the deconvolution of the TPM matrix. For example, quantiseq method provides cell fraction after deconvolution, like that what is the value in ABIS deconvolution. It would be really helpful for me to analysis the result if someone can provide information about this topic.

[MatStat]

Hi, I also not quite sure about how to handle the output of ABIS. They do mention that the values are represented in percentages so summing the values across all types of immune cell should give something close to 100, if the samples are presumably PBMC.

The output values should be percentages of immune cell type, hence within the range of 0-100. Hence, if you use PBMC data, when you sum the scores for each sample, you should ideally get a value close to 100. Note also that the method used is without constraints, hence it is likely that you will obtain negative values. In this case you can consider one of these two situations:

Negative values close to zero -- they are likely due to technical or biological variability. Simply set them to zero or scale all the values up so that the minimum value is zero.
Very low negative values -- there might be strong biological or technical variability for the cell type with such values. Exclude the cell type from the analisys.
  

ref: https://github.com/giannimonaco/ABIS

I encounter two main problems: 1. I don't get any near 100% when summing up that way (as mentioned above). 2. I don't understand how to handle negative results. they do refer to it when explaining the output but not as detailed as I would expect. for example, what are "Negative values close to zero" and "Very low negative values"? what type of normalization is appropriate? min-max normalization? if so what ranges.

Did you manage to solve this? Is there anyone else who knows how to handle it?

All the best

[Barista]

Anyone found a solution yet, I have exactly this problem as of now!