iiinova — where your digital self keeps evolving

Ultimately, our data reflect a clear pattern-- the outcomes of infections are highly heterogeneous, and for herpesviruses, the persistence of the viral genome is driven by MHC II variation. 14/

140 0

Borrowing from the immunotherapy literature, we then scored each individual based on their predicted best peptide to derive a per-allele, per-person score of EBV presentation ability. This measure was strongly predictive for class II alleles and reproducible in both cohorts! 11/

HLA was by far and away the strongest association, so we sought to dig in deeper. In our GWAS, we had the distinct advantage of knowing what peptides were presented on class I&II (i.e., the EBV proteome)--we scored all participants for predicted EBV peptide presentation strength via NetMHCpan. 10/

140 0

Overall, this genetic architecture of EBV DNAemia covaried with various autoimmune conditions, including SLE & RA (known EBV-linked conditions) with an opposite effect for IBD. This demonstrates that GWAS signal for these autoimmune traits is pleiotropic with EBV DNAemia. 9/

1100 0

Next, we used EBV DNAemia as an outcome in a GWAS, identifying 20+ independent loci near many notable immune-associated genes such as HLA, CTLA4, EOMES, etc. The effects were strongly concordant between UKB and All of Us, a replication cohort. 8/

180 0

With EBV DNAemia as a biomarker, we then performed a phenome-wide association study, looking to identify what complex traits are linked to persistence of the viral infection. Many known and novel hits were significant, from respiratory disease to autoimmunity! 7/

By removing these regions, our measure of EBV DNAemia (i.e., persistent DNA in peripheral blood via WGS) could be derived for ~750,000 individuals. For a subset of the UKB where serology was available, our corrected measure was highly concordant only after masking these biased regions. 6/

The workflow wasn't as straightforward as we originally thought. Due to highly repetitive regions of the EBV contig, low-quality alignments gave a pervasive false-positive signal, which we were able to ultimately mitigate. 5/

160 0

In a cool twist of fate, the EBV contig is in hg38 to mop up unscrupulous EBV reads, a by-product of immortalizing lymphoblastic cell lines (used in 1000 Genomes Project, etc.). Hence, a simple `samtools view` could get a measure of persistent EBV DNA in large WGS cohorts, e.g., UK Biobank. 4/

171 0

In this work, we focus on Epstein-Barr virus-- an enigma in many ways. While >90% of us have been infected by this endemic virus, most of us will live just fine. However, for a subset of the population, this infection can be a trigger of disease-- from MS to cancer to autoimmunity. 2/

Epstein Barr Virus and causes of disease

1116 0

Excited to share a new preprint from the lab with @ryandhindsa.bsky.social ! www.biorxiv.org/content/10.1... Led by @sherrynyeo.bsky.social, @erinmayc.bsky.social, and friends, we continue our journey to find viral DNA in our favorite place-- the overlooked and discarded reads in existing data! 1/

the treasure trove of all sequencing datasets

17237 0

PIPseqV uses this "random enzymatic fragmentation sites" to create "intrinsic molecular identifiers". So I think a likely cause isi the barcoding of different parts of the molecule that can be detected vs the strong 3' preference in the 10x workflow.

220 0

To give a flavor for this, here are a few genes almost selectively detected in one technology but not the other. This is the interesting/scary part-- I'm not sure if these genes do anything in immune cells, but we've been blind to them in single-cell and/or there's some false-positive signal /9

320 0

As one would expect, there's a strong correlation in pseudobulk gene expression (global r = 0.90), but what surprised me was many genes were detected almost exclusively in one technology but not the other /8

240 0

The differences start to emerge when you look a bit more closely. In terms of molecular complexity, PIPseq performed pretty well, but there's a pretty massive difference compared to GEM-X. At similar cell numbers and sequencing depth, we get about 63% more UMIs / cell with GEM-X over PIPseq V /7

140 0

Labeling cell types with azimuth, the correspondence between the technologies at this resolution is pretty clear. I dug into the slight differences between the two and couldn't find anything that I trusted (e.g., overclustering / cell quality / mislabels explained anything off the y=x line here) /6

120 0

So what did we see? In brief, both technologies readily delineate the major cell types in PBMCs with clear expression of marker genes we've come to know and love. So at birdseye view, cell state separation and annotation is pretty consistent between technologies. /5

120 0

Caleb Lareau @caleblareau.bsky.social · 08/03/2025

We designed a basic head-to-head experiment using a cryovial of PBMCs from a healthy donor, split and profiled via 10x GEM-X 3' and PIPseq V (T20). @karolisk.bsky.social masterfully generated data on his first try with the Single-cell Analytics Innovation Lab (SAIL) here at MSKCC. /3

120 0

Feeling invigorated after hearing from patients and their families while riding with @cycleforsurvival.bsky.social. I am SO grateful to work with this talented and motivating team at @mskcancercenter.bsky.social! Thank you to everyone who participated in and donated to this outstanding cause!

0224 0

Caleb Lareau @caleblareau.bsky.social · 08/01/2025

We previously hyped up PERFF-seq on the evil bird app as a pre-print. In the fully baked form, we’ve added a ton of new data, including experimental design, integration with conventional protein cytometry, and sorting cells by somatic changes (loss of Y chromosome using an MSY OR gate). 4/n