search

omnideconv / deconvBench

Comparison of 2nd generation deconvolution methods implemented in omnideconv
2 stars 0 forks source link

Preliminary results on real mouse data #8

Closed FFinotello closed 9 months ago

FFinotello commented 2 years ago

Hey @LorenzoMerotto! I think it is a good time to start keeping track on our plans and preliminary results :)

LorenzoMerotto commented 2 years ago

1) Deconvolution of mouse RNAseq data with first generation mouse methods

Tested methods are: mMCPcounter, seqImmuCC, DCQ, BASE

1.1) Petiprez dataset E-MTAB-9271 This dataset consists of RNAseq of 14 samples from 4 different tissue types (spleen, peripheral blood, peritoneum and TC1 grafted tumor), and the corresponding FACS estimates. Some FACS estimates for certain cell types are missing.

immagine

1.2) Wuaiping dataset E-MTAB-6458 This dataset consists of RNAseq of 12 samples from 4 different immune tissues (spleen, bone marrow, lymph node and PBMC), and the corresponding flow cytometry proportions. For this daatset we have less cell types estimated.

immagine

2) Deconvolution of mouse RNAseq data with first generation human immune methods The same dataset was tested with immune based methods for human data, through immunedeconv. Gene names were converted to the orthologs with the BiomaRt package. After the conversion process the size of the gene counts table was halved (55k -> 22k genes). The two counts tables have the same set of genes. This could be the reason behind some poor performances

2.1) Petiprez dataset immagine

2.2) Wuaiping dataset immagine

Aside from problems related to quanTIseq when it comes to the T cells, it is evident that bad results can not be predicted, es EPIC and TIMER estimate well the T/CD8 cells on one dataset and not on the other.

LorenzoMerotto commented 2 years ago

Next step: test the deconvolution methods for mouse on simulated (mouse) data