We (with @dornhaus.bsky.social) finally wrote up our class on how to teach modeling to biologists, which ends up being a non-trivial exercise in understanding the roles that theory plays in the scientific method. We hope others will use our class, or simply enjoy the manuscript!
With Eugene Koonin, we propose a concept of “the selfish ribosome”, under which evolution of life is viewed as a ribosomal takeover, where the ribosome evolved to consume most of the cell’s resources, while other cellular componentry ensures the propagation of the ribosome. arxiv.org/abs/2602.23268
New paper out! Here's a puzzle: phototrophy, the ability to use light for energy, is one of life's great innovations. It evolved early and transformed the biosphere. But it evolved 2x. Why not just once, why not more? Our work suggests the answer is priority effects.
www.nature.com/articles/s44...
They're literally running out of lawyers in a state where the district judges are practically at war with the administration.
A line graph of the number of NIH awards for new and competitive renewal grants made in fiscal year 2026 compared to fiscal years 2021 through 2025. The curve for fiscal year 2026 appears to be essentially at zero on this scale.
New and Competitive renewals
These are still be made VERY slowly. Only 75 awards have been made through 1/23/26.
These are now from 7 institutes and centers (NIA (45), NINDS (16), NIDDK (5), NIDCR (4), NIDCD (2), NHLBI (1), and NCATS (1)).
3/3
🚨 New from me: Grant review at more than half of NIH's institutes could be frozen by the end of the year.
That's because crucial NIH grant-review panels are slated to be empty at those institutes by Jan 2027.
A wonky bureaucratic problem with big implications.
A short 🧵
A new preprint from the lab, with postdoc @deboraycb.bsky.social and collaborators @aidaandres.bsky.social and Tim Connallon:
“Characterising the detectable and invisible fractions of genomic loci under balancing selection”
www.biorxiv.org/content/10.6...
A recent scientific paper compared long-term mortality by vaccination status.
I noticed that Table 2 drew a lot of attention, but was actually included in the paper as a static image. So I built a quick dataviz project to explore.
🧵
Oh no!
Our latest preprint revisits the classic model of mutation-selection balance.
Do human recessive genes fit Haldane's 100-year old model?
This work is by the wonderful @jonj-udd.bsky.social, and co-mentored by @jeffspence.github.io
www.biorxiv.org/content/10.6...
A study on 1.7 million people in Hong Kong shows superior hybrid immunity to Covid in people who got vaccinated before infection vs. people who got infected first. "Our findings are a direct rebuttal to arguments for natural immunity," the authors write. doi.org/10.1016/j.va...
Sequence alignments are notoriously prone to error. Our latest preprint offers a new tool for filtering errors out, assesses it and other filtering tools, and recommends new best practice. www.biorxiv.org/content/10.6... @phylowheeler.bsky.social 1/10
Take-home: substitution matrix quality matters. Instead of using ModelFinder, use IQ-Tree’s QMaker to build your own for your taxon of interest, trained on strictly filtered alignments. And use CLOAK before inferring trees. 10/10
Histograms of the amount that the distance between gene tree and species tree improves or gets worse, as a way of visualizing a paired t-test. The gentlest filter, CLOAK yields the greatest improvement, while strict filters make matters worse.
For the sequence whose gene tree you are inferring, strict filters hurt, and our new gentle filter CLOAK performs best. Propagating uncertainty from our 16 variant alignments into a consensus among 16 variant trees was worse, showing the presence of systematic not just random alignment error. 9/10
Histograms of the amount that the distance between gene tree and species tree improves, as a way of visualizing a paired t-test. The strictest filter, Divvier-pf, yields the greatest improvement.
Using a substitution model trained on strictly filtered alignment data leads to better inference on gene trees, bringing them closer to the known species tree according to Lin-Rajan-Moret distance (an improved extension of Robinson-Foulds distance). 8/10
Box plots showing how the value of post-filtering exchangeability / pre-filtering exchangeability is significantly below 1 for less plausible mutations, and more strongly so for Divvier-pf than for CLOAK.
Stricter filters have stronger effects in reducing exchangeabilities associated with less plausible amino acid substitutions, i.e. those that require more than one mutation, according to the genetic code. 7/10
Box plots showing lower exchangeability values for higher mutational distances, which become still lower after filtering.
Phylogenetics relies on substitution models (rates of evolution between amino acids or nucleotides), which are normally decomposed into a symmetric exchangeability matrix and equilibrium frequencies. Filtering reduces exchangeabilities between amino acids not linked by single mutations 6/10
precision-recall curve confirming that Muscle5 outperforms MAFFT and Clustal Omega, as well as the performances of alignment cleaning programs
In a trade-off between precision and recall, CLOAK is the best gentle filter, the “partial filtering” option within Divvier doi.org/10.1093/molb... is the best strict filter, and TAPER and the Divvier’s divvying option are in between. GUIDANCE2 and HmmCleaner perform less well. 5/10
Our new program CLOAK (CLeaning On Alignment C[K]onsensus) uses 16 variant alignments generated by Muscle5 (3 perturbations of HMM × 3 perturbations of guide tree), and retains only the pairs for which all 16 agree. You can call it within Muscle5 with the option “-cloak”. 4/10
Unfortunately, when building a gene tree, every bit of information can be important enough such that filtering alignments (at least with older tools) can make tree inference worse doi.org/10.1093/sysb... 3/10
highly repetitive region
almost all gaps, with a few isolated aligned residues
isolated region aligning odds and ends that didn’t go anywhere else
Trigger warning: the attached images of real multiple sequence alignments may cause feelings of distress among biologists: 2/10
Sequence alignments are notoriously prone to error. Our latest preprint offers a new tool for filtering errors out, assesses it and other filtering tools, and recommends new best practice. www.biorxiv.org/content/10.6... @phylowheeler.bsky.social 1/10
I wrote about missing heritability, "missing environmentality," and why I still think twin studies are interesting and valuable: kathrynpaigeharden.substack.com/p/twins-are-...
chart of child mortality in history, roughly 48% for all societies across the globe…until the 20th century, where it plummets to 4%
Currently dorking out over this graph about child mortality with my brother. Just mind boggling to take in.
Hi everyone! I'm co-organizing this retreat/workshop June 15-19 for those looking to get started in mathematical/computational modeling of biological processes. Location is a beautiful farm in NC. Please share with students and others who want to build modeling skills. Interdisciplinarity welcome!
I have no idea about whether he is religious. I was responding to what you wrote about how the Nazis would have classified him, which has a clear factual answer.
This article provides more data showing how easy it is to manipulate humans on social media.
These researchers rerouted the algorithm on Twitter to push some users toward “antidemocratic attitudes and partisan animosity”.
It only took 1 week to elicit changes that used to take 3 years.
