[Edit] I think I finally have some winners WITH cell identities, though I had to dig for that part. Thanks!
Anyone working in #Genomics #SingleCell #RNASeq aware of a good data source for human kidney single-cell data that has *labeled* cell types.
We have a #GWAS that the target tissue would be kidneys. I'm stuck doing cell type analyses until I can find some labeled cell type data to connect back to the association data.
#gwas lookup: "Interpret GWAS variant associations across multiple genomics databases"
https://sashagusev.github.io/gwas_lookup/#pos=chr6%3A160540105&build=hg38
SAIGE-GPU — Accelerating Genome- and Phenome-Wide Association Studies using GPUs Open Access
https://academic.oup.com/bioinformatics/advance-article/doi/10.1093/bioinformatics/btag032/8438945
"We developed SAIGE-GPU, a GPU-accelerated version of SAIGE that replaces CPU-intensive matrix operations with GPU-optimized kernels. The core innovation is distributing genetic relationship matrix calculations across GPUs and communication layers. "
Learn the full GWAS workflow in our 5-day online course (26–30 Jan 2026)
@OscarGenomics
From study design to data analysis with R & Linux tools, gain hands-on skills to run your own GWAS.
https://www.physalia-courses.org/courses-workshops/course49/
For once I did not spend much time on biological networks (maybe inadvisable at a course whose subtitle contains the words "network modeling"), but instead presented SMuGLasso, a sparse multitask group lasso for GWAS in diverse populations that we developed with Asma Nouira and that was published just last week. https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1012734
SMuGlasso groups SNPs by LD blocks (as done by Alia Dehman, Christophe Ambroise and Pierre Neuvial in 2015), and treats genetically homogeneous subpopulations of a heterogeneous data set as different but interlinked tasks. An additional sparsity constraints allows us to identify population-specific loci.
On simulations, we show that SMuGLasso has much better recall than classical GWAS (at the expense of more false positives, but fewer than other comparison partners).
On real data, we observe that SMuGLasso recovers all the genes identified by a classical GWAS, as well as some that were identified by a meta-GWAS that included our dataset. All in all, only 2 of the 27 identified genes look likely to be false positives.
Course Alert!😍
Curious about #GWAS? This course will guide you through the entire process — from study design and data prep to statistical analysis and interpreting results.
@OscarGenomics @gwascatalog
https://www.physalia-courses.org/courses-workshops/course49/
"GWAS SVatalog: a visualization tool to aid fine-mapping of GWAS loci with structural variations"
https://www.biorxiv.org/content/10.1101/2025.09.03.674075v1?rss=1
Eli Roberson (he/him)
Pierre Lindenbaum
Physalia-courses
Fabio Barteri 🇵🇸
Chloé Azencott
PLOS Biology
SMT