Excited to share our latest preprint introducing SAIGE-QTL, a powerful tool for single-cell eQTL analysis! Check it out here: https://t.co/53svppNTIt #eQTL #scRNAseq #genomics (1/9)

@madduri Tutorials to run SAIGE-GPU: https://t.co/7dCr8HjEkt Docker and singularity containers for deployment on various cloud infrastructures are available.

Amazing collaborative efforts led by @madduri and Alexis Rodriguez. Applied to run phenome-wide GWASs in MVP, SAIGE-GPU demonstrates a critical framework for leveraging GPU resources to boost the computational efficiency of mixed model approaches in large-scale genetic studies.

A shout out to our amazing team and collaborators, with special thanks to my co-developer @AnnaSECuomo! It’s been such a great pleasure working together! @masakanai @anglixue @TheXavierLab @dgmacarthur @drjosephpowell @dalygene @bmneale We invite your thoughts and feedback! (9/9)

With 100 CPUs, it takes < 2 days to map cis-eQTLs for 20k genes across 14 cell types. Trans-eQTL tests have much lower overhead for reading genotypes. Testing 5.3m variants for 20k genes in 3 cell types with 3625, 82,068, and 463,528 cells costs 7, 19, and 26 hrs (8/9)

It detected 413 trans-eQTLs through genome-wide scans in three immune cell types from OneK1K. (7/9)

It detected 413 trans-eQTLs through genome-wide scans in three immune cell types from OneK1K. (7/9)

SAIGE-QTL allows for multiple user-specified marker-level weights in rare-variant set-based tests. It identified 5,541 eGenes (2,317 unique) with rare/less frequent signals (MAF <= 5%). 483 (21%) are independent from common eQTLs in the same genes (6/9).

https://t.co/uK2FiB0jsW

SAIGE-QTL detected 48.8% more eGenes (i.e., genes with at least one eQTL) compared to TensorQTL (17,218 vs 11,569 eGenes across all 14 cell types). (5/9)

Through simulations and the analysis of OneK1K cohort, including scRNA-seq data from >1.2M immune cells of 982 individuals across 14 cell types, we showed that SAIGE-QTL is well calibrated for both common and rare variants, while outperformed pseudobulk methods. (4/9).

SAIGE-QTL implements the Poisson mixed model efficiently, facilitating genome-wide eQTL scans for millions of cells across diverse cell types. It tests rare genetic variants' effects on gene expression through set-based tests, which are understudied in prior research. (3/9)

Current sc-eQTL methods often rely on the pseudobulk approach or struggle with scalability. Our tool directly models single-cell read counts and captures variability. By leveraging shared information across similar cell profiles, it boosts power to test eQTLs. (2/9)

Would you believe me if I told you that a single variant in a non-coding RNA explains ~0.5% of all undiagnosed individuals with neurodevelopmental disorders (NDD) in @GenomicsEngland ??? I didn’t initially either, but here is the story of RNU4-2 🧵1/9

Excited to share our work in print at @AJHGNews using variant call data to estimate DNA contamination. As our sample sizes get into the millions of genomes, we need methods like this to efficiently process and quality control the data

#PostdocHiring #ResearchOpportunity

🌟 **Join Our Lab https://t.co/K8oVK9dddA as a Postdoc!🌟 Passionate about developing and applying novel statistical methods to study genetics of disease progression and integrating omics data? Explore exciting opportunities and apply now: https://t.co/fmRftCXAWt or email me🧬

Out today @Nature – our analysis of the influences of common nuclear DNA variation on mitochondrial DNA copy number and heteroplasmy across >250,000 people across two biobanks!

@CGM_MGH @harvardmed @broadinstitute @StanleyCenter @GlobalBiobanks @cristenw @SeunggeunL @bmneale @dalygene @XihongLin We are recruiting, so please let me know if you are interested in joining us! @CGM_MGH fosters a deeply collaborative environment, offering abundant prospects for engagement in collaboration and co-mentoring.

@CGM_MGH @harvardmed @broadinstitute @StanleyCenter @GlobalBiobanks @cristenw @SeunggeunL @bmneale @dalygene @XihongLin I am grateful for all the support I received from my fantastic colleagues and friends throughout this process. The number of names is too large and enumerating each would undoubtedly transform this into a substantial narrative on Twitter!