Pinned Tweet
When we design molecule we often want to target properties which are averages for the Boltzmann ensemble, rather than a single configuration or graph.
Yet, current property-guided 3D generative models condition on single conformers.
We target this with Boltzmann-Expected Design
Yet, current property-guided 3D generative models condition on single conformers.
We target this with Boltzmann-Expected Design
4
23
110
5,590
See you in philly!
We're very happy to announce the 2027 Theory & Computation subgroup awardees!! 🎉 @BiophysicalSoc Mid Career: Anđela Šarić; Early Career: @smnlssn ; Postdocs: Chelsea Brown and Mohammad Ahsan 🎉 Sponsored by the Chicago Center for Theoretical Chemistry & PRXlife.
0
1
26
794
Simon Olsson retweeted
We will be conducting a search at the assistant professor level this year for a faculty position at NYU in AI-enabled chemistry. If you know of anyone interested in applying (including yourself), the link to the application is here: apply.interfolio.com/195400
0
5
15
1,398
Implicit Transfer Operators (proceedings.neurips.cc/paper_files/pa…) gets the industry treatment and a significant scale up in the new BioEmu2.
Public preview of BioEmu2! @MSFTResearch AI for Science’s new foundation model for biomolecular structure, dynamics and function. Open weights, MIT license.
github.com/microsoft/bioe…
Movie: BioEmu2-simulated binding of intrinsically disordered peptide PUMA to MCL-1 receptor. 🧵⬇️
0
11
74
4,383
This Thursday we are looking forward to welcome @marwinsegler to the first AI4Science seminar of the fall term! Please see the homepage for more details psolsson.github.io/AI4ScienceSemi…
0
2
9
522
Simon Olsson retweeted
There are still a few spots available!
Happy to announce this years Structured Learning Workshop, Oct 27–28 at Chalmers, Gothenburg! Exceptional speakers from academia & industry. Limited spots, apply to attend via the waiting list: chalmers.ungapped.io/Events/5f08797…
0
4
10
1,614
Simon Olsson retweeted
Catching up on recent arXiv preprints, part 4:
”Simulation-Based Quantum System Inference with Neural Posterior Estimation”, arxiv.org/abs/2609.34995, with @julyhangzou, @rahmlab, and @smnlssn, all at @chalmersuniv, in a joint project between @wacqt_sweden and @WASP_Research.
”Simulation-Based Quantum System Inference with Neural Posterior Estimation”, arxiv.org/abs/2609.34995, with @julyhangzou, @rahmlab, and @smnlssn, all at @chalmersuniv, in a joint project between @wacqt_sweden and @WASP_Research.
3
1
4
259
Simon Olsson retweeted
Come to Chalmers AI Days! 📣 🎉
On Oct 29–30, we celebrate AI research with 4 keynote talks (Volker Sorger, Julian Togelius, Julie Josse, and Demetri Psaltis) and sessions on AI Foundations, AI in Health, AI in Research.
Read more and register here: chalmers.se/en/current/cal…
On Oct 29–30, we celebrate AI research with 4 keynote talks (Volker Sorger, Julian Togelius, Julie Josse, and Demetri Psaltis) and sessions on AI Foundations, AI in Health, AI in Research.
Read more and register here: chalmers.se/en/current/cal…
1
2
9
781
Simon Olsson retweeted
A community-driven strategic roadmap for an atomistic machine-learning ecosystem. Special thanks to @MicheleCeriotti for the massive effort of pulling this together.
arxiv.org/abs/2609.39090
arxiv.org/abs/2609.39090
0
19
105
6,839
Simon Olsson retweeted
Thrilled to share our new preprint: "Simulation-Based Quantum System Inference with Neural Posterior Estimation"!
Check out: scirate.com/arxiv/2609.349…
Check out: scirate.com/arxiv/2609.349…
6
4
6
322
Simon Olsson retweeted
Happy to announce this years Structured Learning Workshop, Oct 27–28 at Chalmers, Gothenburg! Exceptional speakers from academia & industry. Limited spots, apply to attend via the waiting list: chalmers.ungapped.io/Events/5f08797…
0
5
14
3,998
Simon Olsson retweeted
We see this as only an early step in using ensemble information within 3D small mol design pipelines, with many exciting directions from here!
Paper: arxiv.org/abs/2609.15077
Code: github.com/rssrwn/ensembl…
Jointly with @smnlssn, Alessandro Tibo and Jon Paul Janet!
9/N, N=9
Paper: arxiv.org/abs/2609.15077
Code: github.com/rssrwn/ensembl…
Jointly with @smnlssn, Alessandro Tibo and Jon Paul Janet!
9/N, N=9
0
3
14
700
Simon Olsson retweeted
Excited to finally announce a new preprint on Ensemble-Conditioned Molecular Design!
We tackle the issue of incorporating more dynamic structural information into 3D molecular generators by guiding sampling on multiple structural conditions
1/N 🧵
We tackle the issue of incorporating more dynamic structural information into 3D molecular generators by guiding sampling on multiple structural conditions
1/N 🧵
4
11
44
4,472
Simon Olsson retweeted
Ensemble-Conditioned Molecular Design (Irwin, Tibo, Janet, Olsson; AstraZeneca/Chalmers).
Reframes 3D generation as optimization over conformational distributions. Composes vector fields for dual-target binders and active-state-selective agonists.
arxiv.org/abs/2609.15077
Reframes 3D generation as optimization over conformational distributions. Composes vector fields for dual-target binders and active-state-selective agonists.
arxiv.org/abs/2609.15077
2
4
24
1,749
Simon Olsson retweeted
Ensemble-Conditioned Molecular Design
1. The paper reframes molecular design from “find a single bioactive conformer” to “optimize the conformational ensemble”: both (i) ensemble modes (distinct low-energy states such as shapes, pharmacophore arrangements, or pocket-bound poses) rties (aggregate scalars over the ensemble such as mean PSA or mean pairwise RMSD).
2. Core method: ensemble-conditioned guidance for 3D molecular generative models. Instead of training on paired multi-state data, the model is trained on single conditions and multiple conditions are composed at inference by combining the vector fields (continuous coordinates) and log-probabilities (discrete atoms/bonds) produced under each condition.
3. This enables a single trained model to express many design tasks by choosing inference-time compositions: target multiple modes, avoid specific modes (negative weights), mix modalities (shape + pharmacophore + pocket + property targets), and combine arbitrary numbers of conditions—without rebuilding datasets per task.
4. Mode conditions supported: (a) shape profiles represented as noisy coordinate point clouds with tunable fidelity (sigma controls how tightly the generated molecule must match), (b) pharmacophore point clouds with types and optional direction vectors, and (c) “soft” protein pocket conditioning (generate a ligand that fits a pocket without forcing a specific pose).
5. Property conditions supported (demonstrated): mean polar surface area (PSA) across conformers and mean pairwise RMSD as a flexibility proxy. These are provided as target scalar values and integrated via conditioning embeddings, allowing simultaneous control of a bound mode and whole-ensemble behavior.
6. A key technical issue is composing structural conditions from different reference frames. The paper introduces adaptive symmetry learning: encoders are trained to output either equivariant or invariant features depending on whether the input is randomly rotated during training, letting one condition define the generation frame while others are safely composed invariantly.
7. The generator uses joint continuous-discrete flow matching: coordinates are generated by continuous flow matching, while atom types and bond types are generated via discrete flow matching (CTMC). The guidance composition is designed to work consistently across both continuous and discrete outputs.
8. Flexible-size generation is added to pocket-conditioned 3D generation by padding to a maximum atom count (48 heavy atoms) and learning to emit padding tokens. Two training details are highlighted as important: placing pad coordinates at the molecule center-of-mass and aligning padded prior/data atoms with a permutation that minimizes squared distance.
9. Benchmarks introduced: (i) multi-mode conditioning with mode targeting and mode avoidance using compact vs extended conformer shapes; results show strong gains over a size-matched virtual screening baseline, with sigma trading off conditioning fidelity vs chemical novelty. (ii) ensemble property optimization while maintaining pocket + interaction pharmacophore constraints; PSA and RMSD respond monotonically to targets while interaction recovery and docking scores remain largely stable (directional rather than perfectly calibrated control).
10. Case studies: (a) dual-target design for AChE (donepezil) and MAO-B (safinamide) by composing two pharmacophore conditions; docking-based evaluation shows higher fraction of molecules scoring well in both pockets than single-target conditioning. (b) active-state-selective agonist design for A2A adenosine receptor by targeting the active pocket while avoiding the inactive pocket; adding the negative inactive-state condition increases active-state preference in docking (∆Vina state) while keeping low similarity to the known agonist.
📜Paper: arxiv.org/abs/2609.15077
#ComputationalChemistry #Cheminformatics #MolecularGeneration #DiffusionModels #FlowMatching #DrugDiscovery #GenerativeAI #StructureBasedDesign #MultiObjectiveOptimization #ConformerEnsembles
1. The paper reframes molecular design from “find a single bioactive conformer” to “optimize the conformational ensemble”: both (i) ensemble modes (distinct low-energy states such as shapes, pharmacophore arrangements, or pocket-bound poses) rties (aggregate scalars over the ensemble such as mean PSA or mean pairwise RMSD).
2. Core method: ensemble-conditioned guidance for 3D molecular generative models. Instead of training on paired multi-state data, the model is trained on single conditions and multiple conditions are composed at inference by combining the vector fields (continuous coordinates) and log-probabilities (discrete atoms/bonds) produced under each condition.
3. This enables a single trained model to express many design tasks by choosing inference-time compositions: target multiple modes, avoid specific modes (negative weights), mix modalities (shape + pharmacophore + pocket + property targets), and combine arbitrary numbers of conditions—without rebuilding datasets per task.
4. Mode conditions supported: (a) shape profiles represented as noisy coordinate point clouds with tunable fidelity (sigma controls how tightly the generated molecule must match), (b) pharmacophore point clouds with types and optional direction vectors, and (c) “soft” protein pocket conditioning (generate a ligand that fits a pocket without forcing a specific pose).
5. Property conditions supported (demonstrated): mean polar surface area (PSA) across conformers and mean pairwise RMSD as a flexibility proxy. These are provided as target scalar values and integrated via conditioning embeddings, allowing simultaneous control of a bound mode and whole-ensemble behavior.
6. A key technical issue is composing structural conditions from different reference frames. The paper introduces adaptive symmetry learning: encoders are trained to output either equivariant or invariant features depending on whether the input is randomly rotated during training, letting one condition define the generation frame while others are safely composed invariantly.
7. The generator uses joint continuous-discrete flow matching: coordinates are generated by continuous flow matching, while atom types and bond types are generated via discrete flow matching (CTMC). The guidance composition is designed to work consistently across both continuous and discrete outputs.
8. Flexible-size generation is added to pocket-conditioned 3D generation by padding to a maximum atom count (48 heavy atoms) and learning to emit padding tokens. Two training details are highlighted as important: placing pad coordinates at the molecule center-of-mass and aligning padded prior/data atoms with a permutation that minimizes squared distance.
9. Benchmarks introduced: (i) multi-mode conditioning with mode targeting and mode avoidance using compact vs extended conformer shapes; results show strong gains over a size-matched virtual screening baseline, with sigma trading off conditioning fidelity vs chemical novelty. (ii) ensemble property optimization while maintaining pocket + interaction pharmacophore constraints; PSA and RMSD respond monotonically to targets while interaction recovery and docking scores remain largely stable (directional rather than perfectly calibrated control).
10. Case studies: (a) dual-target design for AChE (donepezil) and MAO-B (safinamide) by composing two pharmacophore conditions; docking-based evaluation shows higher fraction of molecules scoring well in both pockets than single-target conditioning. (b) active-state-selective agonist design for A2A adenosine receptor by targeting the active pocket while avoiding the inactive pocket; adding the negative inactive-state condition increases active-state preference in docking (∆Vina state) while keeping low similarity to the known agonist.
📜Paper: arxiv.org/abs/2609.15077
#ComputationalChemistry #Cheminformatics #MolecularGeneration #DiffusionModels #FlowMatching #DrugDiscovery #GenerativeAI #StructureBasedDesign #MultiObjectiveOptimization #ConformerEnsembles
0
4
25
1,812
Simon Olsson retweeted
When we design molecule we often want to target properties which are averages for the Boltzmann ensemble, rather than a single configuration or graph.
Yet, current property-guided 3D generative models condition on single conformers.
We target this with Boltzmann-Expected Design
Yet, current property-guided 3D generative models condition on single conformers.
We target this with Boltzmann-Expected Design
4
23
110
5,590
This work was led by Selma Moqvist, with Richard Beckmann, Ross Irwin and Rocío Mercado supporting. Thanks to Chalmers, WASP and KAW for funding.
Check out the pre-print: arxiv.org/abs/2607.19519
Check out the pre-print: arxiv.org/abs/2607.19519
1
0
7
399