lets say we have this 4 data types single cell RNA seq, single cell TCR seq and Olink & FACS data, so based on this which machine learning model to choose and which model integrates this data and predicts disease type outcome?
•
written
by
Ifthekar Hussain •
1 answer
For the summarisation part the boring options are the ones that hold up: cell type proportions per patient, pseudobulk mean expression per cell type per patient, and on the TCR side clonality metrics - Shannon or Simpson diversity, expanded clone fraction, top-10 clone share. All naturally one number per patient, and it's what most published pipelines actually end up using.
At n=60 I'd worry about this more than model choice: if you pick which cell types or genes to carry forward using all the data and then cross-validate only the classifier, the CV number means nothing. Feature selection has to live inside the CV loop. That leak is behind a lot of implausibly good multiomics AUCs.
I'd also look at MOFA+ before committing to DIABLO. It's unsupervised, so you can squeeze all four blocks down to a handful of factors without the outcome ever touching it, then fit something simple on those. Fewer parameters to burn at your sample size.
Log in to answer this question.
-
News: Webinar “Learning Single-cell Atlases” with Dr. Fabian Theis, author of Scanpy on May 8
written by Claire Watson •
-
Bulk RNA-seq in disease vs controls: deconfound differential gene expression by estimated cell typ…
written by Peter •
-
integration of single cell RNA-seq and bulk RNA-seq data
written by ramyak1912 •
-
RNA-seq dataset search
written by ramyak1912 •
-
Tool: CellKb 2.0: A database of cell type marker gene sets
written by Ashwini Patil •
-
Heterogeneity from bulk ChIP-seq or ATAC-seq data
written by msimmer92
-
Job: Looking a single-cell RNA-seq analysis PhD position /job
written by Paragon •
-
Classification using publicly available RNA-seq dataset
written by bioinfraML •
-
Data for comparison between control and treat samples (cells or patients) by single-cell RNA-Seq
written by landscape95
-
Cutting Edge Single Cell Analysis Methods?
written by informatics bot
Do you realize the magnitude of the answer you're looking for? A colleague of mine has been working on this question for nearly seven years now, and had to design his own model.
Thanks. To make my question more specific, I have a small patient cohort with a binary disease outcome. Olink and FACS data are already at the patient level, while scRNA-seq and scTCR-seq are single-cell data. I am considering DIABLO/mixOmics as the primary method for multi-omics integration and classification, with Random Forest as a secondary prediction model. I am looking for established ways to summarize the single-cell data into patient-level features and would appreciate any references or suggestions for this type of workflow in a small cohort. (approximately 50- 70 patients)