Implement sample sex prediction

[nictru]

Description of feature

Implement sample sex prediction for QC

[StAkinshin]

Hi! We found some existing implementations for sample sex prediction:

1. preprocess&classifySex in Oshlack/speckle (https://github.com/Oshlack/speckle/blob/master/R/; GPL-3 license):
   use combination of 3 params (sum of Y-linked genes count, X-linked genes count, XIST count) for model-based prediction

2. sex_inference_model.R in SexChrLab/SexInference (https://github.com/SexChrLab/SexInference/blob/main/RNAseq/sex_inference_model.R; GPL-3 license):
   R-based method, which predicts sex using pretrained lasso regression model

3. Using Y-chromosome derived genes counts to XIST counts ratio for prediction (https://nbisweden.github.io/workshop-scrnaseq-2024/labs/bioc/bioc_01_qc.html#meta-qc_sex; there is no ready-to-go function for calculation);

4. identify_sex_metadata in kharchenkolab/scITD (line 210, https://github.com/kharchenkolab/scITD/blob/main/R/manage_container.R; haven't found license):
   using counts of RPS4Y1 and XIST for sex prediction

Is it okay to create a Docker container for one of them and implement it to nf-core as a module?

[m-selifanova]

Hi! I'm working with @StAkinshin on sex-prediction functionality. We are planning to add sex prediction module within the QUALITY_CONTROL subworkflow, using the filtered h5ad after DOUBLET_DETECTION and integrating the results into the MultiQC summary. Does this seem like the correct place in the pipeline?

[nictru]

This is another feature that was requested by NBIS and I don't know the details about how they wanted to do this. But it seems like your third point is from the NBIS, so that's probably their approach. This does not mean we have to do it like this, if there is a better method around.

To me it seems that all of these methods are relatively simple and just look at the XIST expression and the last one also the RPS4Y1 expression. Now instead of adding an external dependency that is badly maintained (also R-based tools are always a bit of an extra struggle), I would probably prefer to use something like scanpy.tl.score_genes in a custom script, that incorporates the ideas from the referenced sources (this way there is no need for a new container)

@fasterius any insights from the NBIS side?

[nictru]

@m-selifanova place in the pipeline sounds good