I guess this means I will have to give up remembering PDB IDs of key structures (1ubq, 1fmk, 1xqq, 1shf, 2ci2, 1abd, etc all played key roles in my ph.d.).

Here’s a movie of the dancing queen 💃 7/n

So how does the growth hormone receptor then look? Fortunately my good colleague Birthe Kragelund led an effort to combine, NMR, SAXS, SANS, simulations and more to construct an integrative model of GHR that shows a much more dynamic and interesting molecule than the text book implies. 6/n

That mess is because the entire intracellular part of GHR is intrinsically disordered! Yes, the part of the protein that actually signals after hormone binding does not have a well defined three-dimensional structure. Instead it acts as a scaffold that recruits kinases etc to signal downstream 5/n

You won’t find a detailed structure of full-length GHR in the PDB but AlphaFold gives us this. The extracellular (GH-binding) and transmembrane domains are easily seen in blue indicating medium–high confident predictions. But the rest is a mess and even goes back through the membrane. 4/n

So what is growth hormone and how does it work? If you look in a biochemistry textbook you will find an image such as this GH binds to the growth hormone receptor (GHR) results in receptor dimerization and activation, leading to metabolic changes and growth. 3/n

It’s time for the World Cup final with one of the greatest footballers that may be playing his last cup So time for a short thread on a link between Messi and an intrinsically disordered protein. 1/n

It's nice when you benchmark your own method and find it good. But its even nicer when others benchmark your method and find it good. Figure from: Coarse-grained simulations of long intrinsically disordered proteins: a benchmark of Martini 3 force-fields doi.org/10.64898/202...

Not chemistry, but John Nash’s thesis is also famously short and includes two references (one of which is basically the content of the thesis itself) library.princeton.edu/sites/g/file...

Greetings from the nanodisc field

AF-CALVADOS is now published doi.org/10.1002/pro.... We combine AlphaFold and CALVADOS to simulate flexible multidomain proteins at scale: — Ensembles of >12000 full-length human proteins — Comparison of IDRs alone and I n context for >1500 TFs @sobuelow.bsky.social @kejohansson.bsky.social

In this analysis, papers that have not been posted as a preprints were retracted about twice as often as preprinted papers that were also published in a journal (19 vs 8 pr 10.000 papers). [though the authors note that there may be confounders]

drop a cop

It's nice when you benchmark your own method and find it good. But its even nicer when others benchmark your method and find it good. Table from doi.org/10.1101/2025... benchmarking our AF-CALVADOS approach (doi.org/10.1101/2025...)

Drop me an email 😊 novonordiskfonden.dk/en/grant/rec...

Springer Nature has apparently been unable to contact Professor Planck

Many others have done great work in this space, and we have written a few reviews on these topics in the last years. I expect lots of developments in this area with clever and scalable data+computation. 9/n. n=9 doi.org/10.1016/j.jm... doi.org/10.1016/j.sb... doi.org/10.1016/j.sb...

Again, it's possible to use these simulations together with other sources of data to train generative models that include both the folded regions and more flexible disordered parts. And again to benchmark using NMR and SAXS data (rather than ensembles). 8/n doi.org/10.64898/202...

We can also combine AlphaFold to predict the structured regions with the CALVADOS model for the flexible parts, and the resulting AF-CALVADOS model is pretty good for global conformational properties across the order–disorder continuum, & can be scaled to the proteome. 7/n doi.org/10.1101/2025...

It's also possible to train generative models for conformational ensembles using the simulations (& other sources of data) as input. With people at PepTone and NVIDIA we recently demonstrated that it's possible to include folded domains, and benchmarked using NMR+SAXS. 6/n doi.org/10.1101/2025...

While the simulations are fast and can be scaled to thousands of sequences, they are not as fast as modern ML methods. But we can run simulations at scale, and then train ML models to predict properties directly from sequence. 5/n doi.org/10.1038/s415... doi.org/10.1073/pnas...

We can also "invert" these model to design new sequences with specific conformational properties and validate them using subsequent experiments. The proteins below have the same amino acid composition, but different sequences. The force field extrapolates pretty well. 4/n doi.org/10.1126/scia...

More recently, we applied this to data for a larger number of proteins for which we collected SAXS+NMR-PRE data from the literature to develop the CALVADOS simulation model that we benchmarked across a broader set of experiments on IDPs. 3/n doi.org/10.1073/pnas... doi.org/10.12688/ope...

Some years ago we showed that one can learn the parameters of a force field from experimental data by backpropagating deviations between experiments and simulations from the ensemble to optimize force field parameters parameters directly. Here we used NMR PRE data. 2/n doi.org/10.1529/biop...

The approach we have taken is not to use NMR structures, but instead to optimize a physical model (a coarse-grained force field) using the "raw" NMR and SAXS data, then to run simulations at scale with this model, and then to train ML models using the simulations 1/n Review: doi.org/10.1016/j.sb...

Great. Goes well with this paper on how to write bad papers: doi.org/10.1111/j.00...

Football is a simple game…

Blog post by Kamil Tamiola about our joint paper on On-the-fly Probability Enhanced Sampling in the multithermal ensemble to simulate IDPs idps.substack.com/p/a-disorder... Paper: doi.org/10.1038/s414... @julianstreit.bsky.social @invemichele.bsky.social & Sandro Bottaro

Thanks. Yes it will be interesting to see how much of this turns out to be useful additions to the literature. I’m not too happy about this drawing of isoleucine.

Fixed that for you LEGO