Our new preprint using sex-biased genes for accurate deconvolution of sample identity in single cell transcriptomics is out! This work (led by PhD student Guy Twa and @rphillips3.bsky.social) may be useful for labs that incorporate both sexes in single cell RNA-seq. www.biorxiv.org/content/10.1... - ThreadSky

daylab.bsky.social • 84 days ago

Our new preprint using sex-biased genes for accurate deconvolution of sample identity in single cell transcriptomics is out! This work (led by PhD student Guy Twa and @rphillips3.bsky.social) may be useful for labs that incorporate both sexes in single cell RNA-seq.
https://www.biorxiv.org/content/10.1101/2024.11.29.626066v1

Comments

daylab.bsky.social•84 days ago

The background - single cell transcriptomics is expensive, and poses a challenge in that cost often scales with sample size. Current strategies to deal with this involve combining multiple samples together, followed by deconvolution using hashing barcodes or genotype information.

daylab.bsky.social•84 days ago

However, this requires additional reagents or data modalities. We wondered inherent features of an snRNA-seq dataset (such as genes that are differentially expressed in cells from male and female samples) could be used for deconvolution instead.

daylab.bsky.social•84 days ago

If so, this would allow samples from male and female replicates to be combined for processing but later separated for analysis – effectively doubling the sample size without increasing reagent cost and without adding any new data modalities.

daylab.bsky.social•84 days ago

Here, we benchmarked several potential methods for classifying cell sex using published ground-truth snRNA-seq data, with some interesting surprises and implications.

daylab.bsky.social•84 days ago

Simple models that used expression of Xist or Y chromosome genes alone fail pretty often, and in different ways. Using Xist as a classifier failed most often by misclassifying female cells, whereas Y chromosome gene classifiers failed most often with male cells.

daylab.bsky.social•84 days ago

Machine learning models like neural networks, support vector machine, and random forest do a really nice job of classification when trained on sex-biased gene expression patterns, but also took hours or days to run on our high performance computing system.

Comments

Posting Rules

Reply