iiinova — where your digital self keeps evolving

Reposted by Ari Firestone

bioRxiv Cancer Bio @biorxiv-cancer.bsky.social · 09/05/2026

Dependencies in heterogeneous, lineage plastic patient-derived prostate cancer organoids revealed through integrated single-cell multiomics and CRISPR screening www.biorxiv.org/content/10.64898/20…

012

Ari Firestone @afirestone.bsky.social · 15/04/2026

f you want to quantify individual TE expression alongside genes or transcripts (which I would argue you must) give it a try! t.co/wZAAapaIFR

t.co

https://github.com/calico/majec

100

Ari Firestone @afirestone.bsky.social · 15/04/2026

MAJEC's joint model reduces it to 5%. How? Genes and TE loci compete for reads probabilistically in a shared EM. Runtime: 20 min wall time for 6 BAMs on 6 cores. Huge thanks to the teams amazing postdoc Tian-Yeh Lim for her work on this. Also a shoutout to David Hendrickson

100

Ari Firestone @afirestone.bsky.social · 15/04/2026

We thought this was a minor edge case. It is not. Using Telescope, exon-overlapping TE loci — just 1.1% of all TE features — account for 43% of total TE signal. That's a 40-fold enrichment. Nearly half of what looks like TE expression is actually gene expression leaking through.

110

Ari Firestone @afirestone.bsky.social · 15/04/2026

Genes and TEs shouldn't be quantified separately. ~45% of the human genome is TE-derived, and TE annotations are densely embedded in gene bodies. When you quantify TEs in isolation, you can't tell what's what.

100

Ari Firestone @afirestone.bsky.social · 15/04/2026

So we built one. MAJEC (Momentum Accelerated Junction Enhanced Counting) takes a BAM, a gene GTF, and RepeatMasker annotations, and produces gene, isoform, AND individual TE locus counts faster than the TE tools I was using. doi.org/10.64898/202...

doi.org

100

Ari Firestone @afirestone.bsky.social · 15/04/2026

This started from a very mundane place. I study TE biology alongside gene expression (epigenetic drugs, cancer). Had to use multiple tools and annotation formats. The TE ones were a bit clunky and had challenging dependency. I just wanted one tool that did it all.

100

Ari Firestone @afirestone.bsky.social · 15/04/2026

New preprint! We built MAJEC a tool that jointly quantifies genes, isoforms, and individual transposable element loci from RNA-seq. It replaces 3 tools, runs fast, and uncovered a contamination problem affecting ~40% of locus-level TE signal.

112

Ari Firestone @afirestone.bsky.social · 16/02/2026

This suggests a path forward via rational sequencing and next-gen inhibitors designed against resistant variants, though we also suspect intratumoral heterogeneity in TA-repeat expansions could play a role in limiting response. Big thanks to the entire Calico/AbbVie team for this multi year effort!

000

Ari Firestone @afirestone.bsky.social · 16/02/2026

The silver lining: Resistance is chemotype-specific. Crucially, we found that specific mutations (like C727R and F730L) that confer resistance to the clinical candidate HRO761 were bypassed by our AbbVie/Calico scaffolds.

100

Ari Firestone @afirestone.bsky.social · 16/02/2026

MSI cells are hypermutators that rapidly evolve on-target resistance in the WRN helicase domain (G729, F730, C727, L528). In our models, tumors regrew in weeks and were completely refractory to rechallenge.

100

Ari Firestone @afirestone.bsky.social · 16/02/2026

WRN inhibitors were hyped as the next PARP-like success for MSI cancers, but early clinical returns have been modest. Our new paper, lead by Faith Fowler in MCT is a collaboration between Calico and AbbVie, helps explain why.

100