Lastly, PETRA can be used to assess the performance of state-of-the-art AI models, such as Borzoi and AlphaGenome. Correlations between PETRA scores and model outputs vary highly across the four genes studied, emphasising the need to generate more data in diverse genomic contexts. 7/9

Comparing variant sets between primary T cells and Jurkats reveals highly correlated effects for VAV1, but much less so for IL2RA. 6/9

We define TF motifs that modulate expression and engineer alleles with 2+ motifs via combinatorial PE. Expression effects are largely additive. For example, some alleles with multiple MYBL2 motifs express >10-fold more IL2RA RNA. Conversely, CTCF motifs can effectively silence target genes. 5/9

Scoring of random 6-mer insertions in large libraries is highly reproducible, with dozens of variants having relatively large effects (over 2-fold). Effects of insertions across different genes are largely context dependent, reflecting e.g. transcription factor (TF) binding site creation. 4/9

PETRA affords high scalability and flexibility, as no pre-engineering or elaborate assay optimisation is required. We demonstrate PETRA across four loci in Jurkat cells and two loci in primary T cells, scoring over 14,000 engineered sequences in total.

Highlights of our Results are as follows: 3/9

PETRA leverages prime editing to test variants installed downstream of target genes' transcription start sites. This allows RNA-level effects to be quantified in multiplex via amplicon sequencing of DNA and RNA, akin to the strategy of STARR-seq but with the big advantage of endogenous editing. 2/9

This story is the PhD work of Phoebe Dace, who has done remarkably well to bring this all together. Congrats, Phoebe! 👏
Thanks to the lab, Nicole, @lcubes.bsky.social @chloeterwagne.bsky.social and Megan), our great collaborators, and @crick.ac.uk & @cancerresearchuk.org for vital funding. 🙏 END/16

Rather than confounding clinical interpretation, having data from multiple models is clearly preferable. A path forward to more precisely calibrating the risk caused by individual variants is to integrate functional evidence from multiple experimental data sets generated at scale. 15/n

Finally, we used gold-standard sets of variants to calibrate the strength of evidence provided by scores from each cell line individually and produced ACMG/AMP-style evidence codes, which we anticipate will improve BRCA1 variant classification. 12/n

@mariazanti.bsky.social‬ and @kmichailidou.bsky.social addressed this by estimating breast cancer risk from case-control data. Variants causing loss-of-function ONLY in HAP1 moderately increased risk (OR=3.3). Yet variants that were loss-of-function in BOTH lines highly increased risk (OR>10). 10/n

Importantly, both the HAP1 and HMEC SGE assays perform well for predicting which BRCA1 variants are pathogenic. Looking at a set of 420 pathogenic and benign variants in ClinVar, it was the HAP1-based assay that performed better overall, with near-perfect accuracy. 8/n

To better understand why variants act differently between lines, Phoebe performed more SGE using PARP inhibition and a knock-out HMEC line to reveal context-specific effects. Many variants initially scoring discordantly between HAP1 and HMEC can be reconciled in specific contexts. 7/n

Comparing how variants impact function across cell lines has proven quite interesting. Nearly half of variants leading to loss-of-function in HAP1 do not have the same effect when introduced to HMECs. Variants with discordant effects are typically missense and splice. Here's an example... 6/n

She also engineered a new SGE assay in human mammary epithelial cells (HMECs). 5/n

In each of these key studies, effects of variants were measured in only a single cell line.

This is what makes Phoebe’s new work so impressive. She not only tested over 4,000 new BRCA1 variants in our optimised HAP1 system… 4/n

In @jshendure.bsky.social lab in 2018, we performed saturation genome editing (SGE) to test 3,893 genetic variants in BRCA1. SGE is a CRISPR method to engineer variants in lab-grown human cells. This work showed that SGE could be used to identify variants that cause high cancer risk in people. 2/n

We're hiring again! Now looking for a postdoctoral fellow to join us on our mission to improve methods for testing human variants at scale. 🔎🧬🧪

Many potential projects to tackle, all in the top-tier research environment of the @crick.ac.uk.

crick.wd3.myworkdayjobs.com/External/job...

Huge congratulations to Magdalena Armas @magdaarmas.bsky.social (PhD Student @crick.ac.uk) for winning the Silver Award at the STEM for Britain competition!