Hi, I also have the same problem. Did you find a solution?

Hi, at the moment I don't normalize my Raw Counts matrix, as it's mentionned on Cibersortx tutorial (on it's website), raw counts are recommended. Don't forget that to get the bests results as possible, your signature matrix and your Bulk RNA-Seq must be in the same space normalization (in my case I use Raw Counts). I will not recommand RPKM to perform cellular deconvolution even if Cibersortx is able to convert it into TPM.

CIBERSORTx will automatically normalize the input data such that the sum of all normalized reads are the same for each transcriptome. If a gene length-normalized expression matrix is provided (e.g., RPKM), then the signature matrix will be in TPM (transcripts per million). If a count matrix is provided, the signature matrix will be in CPM (counts per million). Regardless of the input, the signature matrix and mixture files should be represented in the same normalization space.

Read this excellent article to know more about that :

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7648640/

> Did you make any filtration on raw data prior creation of the signature matrix?

No more than the filtrations I made in Seurat to process the data.

Hi, here is the code, it's way too long for a simple thing but it has the advantage to be fast and the ram consumption is very low compared to pandas or R.

# This script takes the output file created in seurat and process
# it to be taken by Cibersortx to create signature matrix
# The file generated by this script has cell type per column
# and raw count for gene in the intersection of a cell type and a 
# gene.
import click # pip install click

# Dictionary
dict_sequence_label = {}
# Key = cell sequence ; value : cell label 

# Opening file with the conversion between sequence and cell label
# Column 1 : Single-Cell Barcode; Column 2 : Label (CD4, B, CD8 etc...)
input_directory = click.prompt('> Enter absolute path of input files directory')
output_directory = click.prompt('> Enter absolute path of output directory')

with open(input_directory+"Convert_UMI_Label.tsv","r") as file_cell_label:
    next(file_cell_label)
    #useless first line
    for line in file_cell_label:
        clean_line = line.rstrip("\n").split("\t")
        sequence = clean_line[0].replace("-",".") #Store barcode
        cell_label = clean_line[1] # Store Cell Label
        dict_sequence_label[sequence] = cell_label

print("dictionnary done !")

header = []
# Now i got the conversion between sequence and cell type
# i'll read the file with gene count per cell sequence
# and write a new file with the dictionnary above
with open(input_directory+"Gene_Count_Per_Cell.tsv","r") as file_gene_count:
    first_line_cell_sequence = file_gene_count.readline()
    #The first line of export in R present a mistake 
    #i save it and correct after
    #next(file_gene_count) #Delete first line
    with open(output_directory+"preSigMatrix.tsv","w") as file_matrix:
        split_first_line_cell_sequence = first_line_cell_sequence.rstrip("\n").split("\t")
        for element in split_first_line_cell_sequence:
            header.append(dict_sequence_label[element])
        file_matrix.write("GENES\t"+'\t'.join(header)+"\n")
        # Once I wrote all the cells label I can write the gene counts
        for line in file_gene_count:
            file_matrix.write(line)
# The script in finished and the file is now created.

print("Single Cell Raw Counts Annotated Matrix susccessfully created!")