Grant Rotskoff

I'm hiring a postdoc with a flexible start date (any time in 2026). Come work with us on topics at the interface of machine learning, biophysics, a nonequilibrium statistical mechanics. If interested, send me a CV and a short summary of why you think you'd be a good fit.
statmech.stanford.edu

Small angel X-ray scattering

Big fan of this perspective:

The plan at FutureHouse has been to build scientific agents for discoveries. We’ve spent the last year researching the best way to make agents. We’ve made a ton of progress and now we’ve engineered them to be used at scale, by anyone. Free and on API.

What an incredibly cool paper! While knot theory strictly applies to closed curves, Tommy, @smnlssn.bsky.social , and @paulrobustelli.bsky.social show that writhe, a knot "non-invariant" that changes with smooth deformations, provides a meaningful descriptor for flexible conformations.

Figure from the paper illustrating sequence–ensemble–function relationships for disordered proteins. ML prediction (black) and design (orange) approaches are highlighted on the connecting arrows. Prediction of properties/functions from sequence (or vice versa, design) can include biophysics approaches via structural ensembles, or bioinformatics approaches via other hetero- geneous sources. The lower panels show examples of properties and functions of IDRs for predictions or design targets. ML, machine learning; IDRs, intrinsically disordered proteins and regions.

Our review on machine learning methods to study sequence–ensemble–function relationships in disordered proteins is now out in COSB

authors.elsevier.com/sd/article/S...
Led by @sobuelow.bsky.social and Giulio Tesei

Amazingly, this trick works. Due to improved algorithms for learning the score coming from the generative modeling, the applicability of this approach is very broad. I had spent several years making false starts on implementing the Malliavin calculus, but in the end, we found a route around it ;)

Jérémie Klinger found a simple trick back to Girsanov: you take the perturbation in the diffusion at the level of the Fokker-Planck equation and rewrite it to be included in the drift. The resulting drift then has a term proportional to \nabla \log \rho(x,t), what machine learners call the score!

The “classical” strategy for diffusion sensitivities comes from financial mathematics and is called the Malliavin calculus, it’s very explicit for simple models like Black Scholes but for a general Langevin equation, it is no easy feat to compute the sensitivity.

In many biological and active systems, diffusivity is highly spatially dependent, and the theory for perturbations in such cases is rather limited, largely based on beautiful work by Leticia Cugliandolo and work by Falasco and Baeisi, among many others.

Far from equilibrium, it is not so easy: one needs to understand the dynamics, and this requires working with dynamical trajectories and their associated path measures. Classically, we do this using the Girsanov theorem, which constructs a “relative path measure” as we perturb the drift term.

Computing response functions or “sensitives” requires understanding how an external perturbation drives the change in some observable. For equilibrium systems, Onsager taught us that this can be understood with correlation functions.

Excited to see our paper “Computing Nonequilibrium Responses with Score-Shifted Stochastic Differential Equations” in Physical Review Letters this morning as an Editor’s Suggestion! We uses ideas from generative modeling to unravel a rather technical problem. 🧵 journals.aps.org/prl/abstract...

Applications for the FutureHouse Independent Postdoctoral Fellowship are due in two weeks! $125k annual stipend, full access to our resources, be coadvised by world class professors and apply our AI science agents to make new discoveries. Apply!

Details here: www.futurehouse.org/fellowship

Ten simple rules for developing good reading habits during graduate school and beyond

To me, the most important are:
Read often, read broadly (incl. older papers and outside your field), and learn to read some papers in detail and others more superficially (and quickly)

Protein-Based Degraders: From Chemical Biology Tools to Neo-Therapeutics The nascent field of targeted protein degradation (TPD) could revolutionize biomedicine due to the ability of degrader molecules to selectively modulate disease-relevant proteins. A key limitation to ...

Excited to share this beast of a review on the potential of protein-based degraders from trainees who are all not on the Sky! Herein we cover choice of binders, selection strategies, E3 ligases, and DELIVERY! pubs.acs.org/doi/10.1021/...

I am hiring a postdoctoral scholar with a start date summer or fall 2025. Projects will be focused on thermodynamically consistent generative models, broadly defined. If you’re interested, please send a CV and one paragraph about why you think you’d be a good fit to rotskoff@stanford.edu

Lot of cool stuff in here. Consistent with my working hypothesis that the main scientific utility of LLMs at the moment is plain old NLP

Really cool opportunity via futurehouse. Come work with them and collaborate with us at Stanford!

If you didn't see our poster at NeurIPS on how to make diffusion model inference fast, you can always read the paper here: arxiv.org/abs/2405.15986

D3R grand challenge 4: blind prediction of protein-ligand poses, affinity rankings, and relative binding free energies - PubMed The Drug Design Data Resource (D3R) aims to identify best practice methods for computer aided drug design through blinded ligand pose prediction and affinity challenges. Herein, we report on the resul...

also I must say often when I read new methods being pre-printed, while I appreciate the eagerness to make a splash, many folks seem unaware of the long history of this field & its assessments - to their detriment

If in CADD, pls read through D3R's last paper
pubmed.ncbi.nlm.nih.gov/31974851/

Can verify that the code works, too :)

@franknoe.bsky.social presented this very impressive work at a fantastic @cecamevents.bsky.social workshop this week. I’m very excited to take a deep dive into the details this weekend!

If you're at NeurIPS next week come see our spotlight poster led by Yinuo Ren and Haoxuan Chen! We use the parallel sampling technique to rigorously establish a big acceleration for diffusion model inference! neurips.cc/virtual/2024...

The large deviation approach to statistical mechanics The theory of large deviations is concerned with the exponential decay of probabilities of large fluctuations in random systems. These probabilities are important in many fields of study, including st...

If you're talking about order in the dynamics itself, then dynamical large deviation theory is very relevant and the canonical review is by Hugo Touchette arxiv.org/abs/0804.0327

There's no easy way to do this in general, but computing the stationary distribution for a nonequilibrium dynamics might be a possible in some low dimensional system or systems with special structure ( ones where you can represent the distribution with tensor networks). Simulations otherwise...

Thanks for looping me in @erikhthiede.bsky.social If by order, you mean that there's some parameter that stabilizes to stationary or periodic steady state, then there the most general solution is simply solving for the stationary distribution or long time limit of the expectation of the order param

Accurate and Efficient Structure Elucidation from Routine One-Dimensional NMR Spectra Using Multitask Machine Learning Rapid determination of molecular structures can greatly accelerate workflows across many chemical disciplines. However, elucidating structure using only one-dimensional (1D) NMR spectra, the most readily accessible data, remains an extremely challenging problem because of the combinatorial explosion of the number of possible molecules as the number of constituent atoms is increased. Here, we introduce a multitask machine learning framework that predicts the molecular structure (formula and connectivity) of an unknown compound solely based on its 1D 1H and/or 13C NMR spectra. First, we show how a transformer architecture can be constructed to efficiently solve the task, traditionally performed by chemists, of assembling large numbers of molecular fragments into molecular structures. Integrating this capability with a convolutional neural network, we build an end-to-end model for predicting structure from spectra that is fast and accurate. We demonstrate the effectiveness of this framework on molecules with up to 19 heavy (non-hydrogen) atoms, a size for which there are trillions of possible structures. Without relying on any prior chemical knowledge such as the molecular formula, we show that our approach predicts the exact molecule 69.6% of the time within the first 15 predictions, reducing the search space by up to 11 orders of magnitude.

Chemists use NMR spectroscopy to identify molecules, but interpreting spectra is laborious and error prone. We show the process can be automated end-to-end using a well-designed Molecular GPT. Importantly, we also make predictions of substructures for interpretability. pubs.acs.org/doi/10.1021/...