Finally, I have to shout out everyone who made this possible, especially the endlessly patient @rachellegaudet.bsky.social but also my fearless mentee Camille and far more people than I can fit in this post who gave me neverending help and advice!
Why would this happen? At the end, we propose a simple model explaining why the same residues might influence both epistasis and specificity in terms of the energetic balance between the transporter's major conformations. Check out the paper and let me know what you think!
Second, we find only a small set of "core" mutations clustered around a key binding site methionine allow for Mg2+ import. However, many additional mutations can then finetune this specificity. What's more, these specificity modulator mutations correlate strongly with those epistatic hotspots (!)
First, even though many mutations combine additively, we see specific epistasis at non-contacting positions that clusters at epistatic hotspots around where the transporter protein opens and closes. We see much more of this than has been observed previously for folding or binding domains.
If you've read this kind of study before, some trends are obvious by now: mutations to the binding site tend to be deleterious, P/R/K are especially perturbative, the effects of many mutations when combined can be modeled as additive. But where it gets interesting is where those trends break down...
To test more than single-mutant effects, we used structural and evolutionary information to guide the library toward combinations of mutations hypothesized to be more likely to alter specificity. This included a combinatorial library across 59 positions informed by natural sequence diversity:
We addressed this by developing new assays that allowed us to make thousands of different sequence changes to a model metal transporter and then quantify how those changes affect its import of a representative native metal substrate, Mn2+, as well as a non-transported ion, Mg2+.
Transporters must use a limited set of structural scaffolds to accurately discriminate between arbitrary substrates, in the process undergoing major conformational change. Despite advances in machine learning, the "rules" by which transporter sequences encode these functions remain opaque.
I'm excited to share our new preprint representing the bulk of my PhD work, along with @camillefreedman.bsky.social, @deboramarks.bsky.social and @rachellegaudet.bsky.social. How do transporter proteins control what molecules they bring across the membrane?
www.biorxiv.org/content/10.6...
First, even though many mutations combine additively, we see significant specific epistasis at non-contacting positions that cluster at hotspots around where the transporter protein opens and closes. We see much more of this than has been observed for folding or binding domains.
If you've read this kind of study before, some trends are obvious by now: mutations to the binding site tend to be deleterious, P/R/K are especially perturbative, the effects of many mutations when combined can be modeled as additive. But it gets interesting where those trends break down...
To probe beyond single mutants and understand how mutations combine, we used both structural and evolutionary information to guide our library - including a particularly fun combinatorial library design informed by natural sequence variation enriched for residues more likely to alter specificity
New paper “Proteome-wide model for human disease genetics” is now live at Nature Genetics: rdcu.be/eRu7K
popEVE (pop.evemodel.org) finds the needles in the haystacks of human genetic variation:
I’m voting no on the continuing resolution that would double healthcare premiums for 20 million Americans, kick 15 million people off Medicaid & allow 50,000 Americans to die unnecessarily every year.
All to give $1 trillion in tax breaks for billionaires.
1/14
Happy (but mostly relieved) to share my dream project with @boudkerlab.bsky.social, now published in @natsmb.nature.com! We used evolution, protein engineering & cryoEM to uncover how ion coupling in glutamate transporters works, and how it evolved.🧵
Free article: go.nature.com/4oRUC1q
Yes, I think I agree. There are two problems: 1) the approach they use isn’t very accurate even for natural proteins and 2) any model will behave very differently for sequences close to the training set than for those far away, so their result is expected just from how the model is trained
To my knowledge there is no evidence that that networks like ESMFold accurately predict structural perturbations upon mutation? So yeah seems it’s entirely analysis *of the model* not of actual protein biophysics
RFK Jr. is not a vaccine skeptic. He's a vaccine denier whose anti-science extremist views and actions are destroying one of humanity's greatest scientific achievements. He's going to kill a lot of people, and those deaths are on the 52 Republican senators who confirmed him and those enabling him.
This is a very cool ancestral reconstruction study by @krishnareddy.bsky.social et al. that I recommend reading! @rachellegaudet.bsky.social and I thought it was so interesting that we wrote a News & Views about it, check it out: rdcu.be/eCfyl
🚨New paper 🚨
Can protein language models help us fight viral outbreaks? Not yet. Here’s why 🧵👇
1/12
🚨 New paper 🚨 RNA modeling just got its own Gym! 🏋️ Introducing RNAGym, large-scale benchmarks for RNA fitness and structure prediction.
🧵 1/9
As a note - AlphaFold and other similar algorithms can still do reasonably well without MSA info. They really have learned principles from the PDB (but these principles are NOT physical protein folding - as evidenced by that they can’t predict destabilizing muts etc)
Couldn't be prouder of my incredibly talented mentee Camille and glad to see her hard work has been acknowledged by winning *both* the Hoopes and Henderson prizes for her undergraduate thesis!
(1/7) Very excited to share my first PhD preprint on the interactions of two of my favorite mobile genetic elements: phages and group II introns!
www.biorxiv.org/content/10.1...
Lovely article from our colleague Kseniia Petrova, who has been wrongfully imprisoned for months: www.nytimes.com/2025/05/13/o...
Thanks, will check these out!
Do you think it would be be more reasonable to benchmark these kinds of models against raw NMR data (e.g. by simulating NMR data from the predicted ensemble, if that’s possible) or are there still too many sources of variability going into the raw data?
Super excited to share a new preprint from our lab on design of small-molecule binding proteins using neural networks! The paper has a bit of everything. A new graph neural network, new design algorithms, and experimental validation. www.biorxiv.org/content/10.1...
🧵🧪
