


Landing soon: An academic consortium for open-source, statistical structural biology!


5/5 So rather than asking whether a dataset “contains an ensemble,” we suggest thinking in terms of a new research frontier: ** ensemble resolution ** Specifically, what differences between distributions can your data distinguish? What do you think? Let us know!

4/5 Then comes processing. Cryo-EM preserves molecule-to-molecule variation, but filtering, alignment, classification, and reconstruction can remove or compress that information. Crucially, a reconstruction can gain spatial resolution while losing ensemble information.

3/5 Single-particle measurements preserve variation more directly. But neither case gives you the ensemble for free. Averaged data constrain only certain features of the distribution. In single-particle, noise, nuisance variables, and finite sampling interfere.

2/5 Different experiments encode the distributions in very different ways. An averaged measurement like X-ray diffraction can still contain a distribution: averaging observables is not the same as averaging structures.

1/5 Ensembles are hot in structural biology! But here’s a question we don’t ask enough: What information survives measurement and processing? In our latest blog, @PilarCossio2 @EmilThomasen and @mhli41 have a think 🤔: rs-station.github.io/2026/08…

5/5 So rather than asking whether a dataset “contains an ensemble,” we suggest thinking in terms of a new research frontier: ** ensemble resolution ** Specifically, what differences between distributions can your data distinguish? What do you think? Let us know!

4/5 Then comes processing. Cryo-EM preserves molecule-to-molecule variation, but filtering, alignment, classification, and reconstruction can remove or compress that information. Crucially, a reconstruction can gain spatial resolution while losing ensemble information.

3/5 Single-particle measurements preserve variation more directly. But neither case gives you the ensemble for free. Averaged data constrain only certain features of the distribution. In single-particle, noise, nuisance variables, and finite sampling interfere.

OpenBind intends to collect 10,000s of protein-ligand structures & affinities. To prioritize what we collect next, we need cofolding models trained on the latest data. Today we're releasing OpenBind-0 and 717 new ligand-bound structures.

7/ Where we're going: teaming up with @open_fold, OpenBind, and cryo-EM + X-ray facilities to make structure beyond the single-structure frontier a reality. Want in? We’re an open community of developers. Join us: 🔗 rs-station.github.io/ slack: reciprocalspa-cfs9659.slack.…

6/ Concretely, we will build: • Data libraries for diffraction, scattering, imaging, and time-resolved experiments. • Differentiable forward models that connect experiments to conformations. • Data-driven tools that turn experiment into structure imbued with dynamics.

5/ To address that gap, from our new home at OMSF we're building the AI↔️structure interface. It cuts both ways: • AI to analyze experiments (new&old) • Exp. observables will train the next bioAI Downstream: cofolding, protein-protein interfaces & biomolecular-dynamics design

4/ AlphaFold and peers were trained on deposited structures, not experimental observables. With this compression, important biology, like dynamics, got lost along the way.

2/ What does that mean in practice? We believe Structural Biology Has More to Teach AI Than Coordinates: 🔗 rs-station.github.io/2026/07…

This is the future, the way real proteins work to process information and energy -- we are looking forward to seeing experimental (wet-lab) validation of the designs!
Binder design has come of age thanks to generative models—but how can we access the wider array of dynamic, multistate protein functions, so elegantly employed by nature? @mihirbafna14 and I are excited to share SwitchCraft, a framework for designing such functions. (1/7)