iiinova — where your digital self keeps evolving

@nahuja.bsky.social also sequenced 20 additional Nanomia from around the world. Population genomics supports three previously described species plus at least one additional undescribed species, with some species occurring in overlapping geographic regions.

020

Key developmental gene families, including homeobox and Wnt genes, have not undergone major expansions. And genes expressed in the same zooids are generally not clustered together in the genome

Does that unusual genome organization explain the extraordinary division of labor among zooids? We looked for some obvious possibilities and there is no smoking gun.

The Nanomia septata genome is ~1.7 Gb on just 8 chromosomes. Closely related non-siphonophore hydrozoans have 15 chromosomes, indicating extensive fusion, translocation, and genome reorganization during siphonophore evolution.

Siphonophores are colonial animals- Like a coral, they have many genetically identical bodies that grow from same embryo. But siphonophore bodies are specialized for different tasks and precisely organized. Many open questions about how this unique colony-level organization evolved and develops.

Our new paper in PLoS One on siphonophore genomes, with chromosome-scale genome for Nanomia septata! Led by @nahuja.bsky.social with @dalila-de.bsky.social, @shchurch.bsky.social, @dts.bsky.social and others. doi.org/10.1371/jour...

1258

Totally- can ask it to show all evidence. Just references, or references with text that contains relevant evidence

000

Let us know how you use it, and please provide feedback on the github issue tracker.

011

There is no strict criterion for what a relevant pdf is, you could put anything you want in (eg a giant folder of mollusc development papers), but most of our optimization has been around handling the challenges that arise when analyzing systematic/ morphology/ taxonomy papers.

111

Processing is roughly O(n), we have used it with less than a hundred pdfs, results improve the more you have. Developed it against libraries that are ~2,000 pdfs and ~20,000 pdfs

100

Heavy processing (language detection, OCR, text extraction, figure segmentation, caption binding, chunking, vector embedding, etc) happens once. pdf processing is idempotent - add new pdfs to library, hit run, and only new files are processed.

200

All scanned pdfs are re-OCRed. It handles pdfs spanning centuries of typesetting conventions (the oldest pdfs we have used it on are for documents from the 16th century). Original language is preserved, LLM handles translation, supporting of multilingual document sets.

100

There are multiple optional constrained ontologies that improve performance, including a recommended bibliography file with reference data for the pdfs and DwC-A files with taxonomic data.

100

"I am new to the study of Hydrozoa. Write a summary about the diversity of the group, the history of its study, and major open questions about their biology. Make it accessible to someone with no expertise in Hydrozoa"

"Create a character matrix for all the most commonly scored traits of siphonophores and generate a tsv table of data I can map onto a phylogeny". It will identify which traits are scored most consistently across species, extract the relevant data from across papers, and generate the table.

100

"Make a single figure that shows bracts and nectophores for all species of Lensia" It will then search for mentions of Lensia in figure captions, grab relevant figure regions across multiple papers, and organize them all in one plot.

100

"I am collecting siphonophores in the surface waters of Saint Helena. What species can I expect to see? How can I tell them apart?" It will then search for species expected in the area, identify diagnostic traits, and streamline to those relevant at your location.

100

You can ask simple questions like...

100

Casey Dunn @caseywdunn.bsky.social · 05/10/2026

Corpus is a clade-agnostic tool. Point it at a folder of pdfs about your clade of interest, it produces a corpuscle - a bundled mcp server for your clade that you then install in a LLM (eg codex or claude). All the information in the pdf library is then searchable and available to the LLM.

210

Interested in joining our lab (dunnlab.org) and YIBS as a postdoc? I'm looking for someone to work on siphonophores, diversity in the open ocean, or natural history. If that's you, please reach out this week to discuss a nomination for the Donnelley Postdoc Fellowship - yibs.yale.edu/donnelley-fe...

dunnlab.org

The Dunn Lab

Casey Dunn's laboratory in the Department of Ecology and Evolutionary Biology at Yale University.

02021

Reposted by Casey Dunn

Pleuni Pennings @pleunipennings.bsky.social · 29/09/2026

I am hiring a postdoc! Are you interested to work on population genetics/ drug resistance/ data? Want to work in Montpellier, France? Fastq files don't scare you? PhD required, start date hopefully Nov 1! (end date Sep 30 2028) Please share! abetterscientist.wordpress.com/2026/09/29/p...

abetterscientist.wordpress.com

Postdoc position on drug resistance evolution in Montpellier, France

I have some exciting news (again!). I got money from the region (Occitanie) to hire a postdoc (and later an engineer / technician) to work on drug resistance evolution in malaria (P. faciparum) and…

179105

Casey Dunn @caseywdunn.bsky.social · 08/09/2026

New paper with @jacobmusser.bsky.social @dts.bsky.social on reproducibility of LLM assisted research. Much of what we learned about engineering reproducible work in past decades still applies. Central organizing principle is whether the LLM sits on data path or off it. www.cell.com/patterns/ful...

085

Casey Dunn @caseywdunn.bsky.social · 31/08/2026

That is so exciting to hear! Please let me know if you have any feedback.

Casey Dunn @caseywdunn.bsky.social · 31/08/2026

Thanks! Art is by the amazing James Prosek - www.jamesprosek.com

020

Casey Dunn @caseywdunn.bsky.social · 31/08/2026

September 4! Not sure how long it will take for stores there to pick it up in their systems.

MBARI @mbarinews.bsky.social · 28/08/2026

Reposted by Casey Dunn

MBARI hosted experts from around the world for Ctenopalooza Ctwo. Researchers shared discoveries and sparked exciting collaborations, while @montereybayaquarium.org shared new developments in culturing these dazzling drifters. 🌊🌈🌊 Learn more: www.mbari.org/news/mbari-h...

mbari.org

MBARI hosts workshop to share latest research on comb jellies • MBARI

This week, MBARI welcomed experts from a wide range of scientific disciplines to discuss the latest research on ctenophores, also known as comb jellies. The Ctenopalooza Ctwo workshop helped foster ne...

1286

Ug, thanks for heads up. Their site isn't so great - after entering address need to click "Update order" for payment section to respond. Or maybe there is another issue too... Super glad you were able to get it from Powell's (one of my favorite bookstores - my grandpa used to take me there as a kid)

010

Thanks to students, TFs, colleagues, and everyone who provided feedback along the way.

060

And I self published the book so I can distribute it online for free and keep physical copies as cheap as possible ($29.99 for the paperback).

1120

I wrote this book for my Phylogenetic Biology course at @yale.edu (EEB 354). There is intentionally one chapter per week of class.

140

I also emphasize the conceptual unity of simulation, inference, comparative analyses, ancestral-state reconstruction, and other methods. They use the same machinery, differing mainly in what is fixed and what is estimated. Like one equation where you sometimes solve for x and sometimes for y.

140

Evolutionary models are generative: they describe how organisms evolve from starting states. But they’re often taught first through inference, like learning to drive in reverse before driving forward. I start with simulation, where they are more intuitive, then extend to inference and other uses.

1101

Alex de Mendoza @alexdemendoza.bsky.social · 03/08/2026

My new book, Phylogenetic Biology, is now published. Available online for free at dunnlab.org/phylogenetic... . Physical copies can be purchased at shop.lightningsource.com/b/085?params... or your favorite bookstore.

12324164

Reposted by Casey Dunn

Sailing in @biorxivpreprint.bsky.social ready for summer, our study on the iconic Portuguese man o' war 🪼 is out: www.biorxiv.org/content/10.6.... We use (epi)genomics to understand a critter that fascinates me since childhood. Massive team effort with @obog.bsky.social, Cummins and Neely groups. 🧵

the indo-pacific bluebottle (Physalia utriculus)

810355

It is now the first analysis we run on new data. We use it for QC (verifying species ids and testing for contamination). We then drop sPCR products into a phylogeny to see where our new samples fit in with previous data. Less than an hour after data delivery we already know a lot about our samples.

Because it uses PCR primers, the regions it assembles correspond to the best sampled gene regions in public archives. This provides a bridge between raw genomic data and decades of work sequencing specific PCR amplified genes from a broad diversity of organisms.

100

Here, for example, is a one-liner that downloads 1M reads from SRA accession SRR23143286 (the siphonophore Nanomia bijuga), runs sPCR with the cnidarian primer panel, and dumps the amplified products to the terminal. It took 41 seconds to run on my laptop, including data download.

100

You can work with local raw data, or give it an SRA or ENA accession number and it downloads (and caches) the data directly. And it only downloads the number of reads you ask it to analyze, so no long waits for giant files that fill your disk.

100

Depending on the gene, genome, and primers, reliable assembly can take as few as 1M reads. High copy genes, like mitochondrial genes and rRNA are often robust with a small number of reads. Single copy nuclear genes take more data and work best with specific (non-degenerate) primers.

You can specify primer sequences at command line, or in a yaml primer panel file. It comes pre-loaded with panels for a few clades. Users can develop and optimize their own primer panels, and submit for inclusion in sharkmer. Let is know what clades and genes you would like to have primer panels for

100