Linna An (@alchemist_an) header
Linna An (@alchemist_an) profile picture

TTAP @RiceUniversity | computational biochemist | Protein designer @UWproteindesign & natural product biochemist @ChemistryUIUC | open to industry

Joined August 2022
Photo by Linna An (@alchemist_an)Photo by Linna An (@alchemist_an)Photo by Linna An (@alchemist_an)Photo by Linna An (@alchemist_an)Photo by Linna An (@alchemist_an)Photo by Linna An (@alchemist_an)Photo by Linna An (@alchemist_an)
Super cool work. Would be great if the discovery can benefit all who are trying to use discovered knowledge to do something useful.
Biologists use specialized open-source models for tasks like modeling the structure of molecular systems, designing drug-like molecules, and predicting the effects of genetic mutations. But these models are often expensive to run, potentially limiting their impact. In our latest Science Blog, we share how Claude was able to optimize inference for more than 30 open-source models, making them 4x faster on average, partly by writing custom software for GPUs. We’re open sourcing all of the optimization code. Read more: anthropic.com/research/claud…
1
6
665
This week I had the honor of speaking to Princeton’s entire incoming undergraduate class to address their AI anxieties. I had three messages for them — good news, bad news, and a note of optimism. Here’s a condensed version. The good news We have enough evidence now to conclude that the shrill predictions of rapid, massive job loss were misplaced. Even in a field like software engineering where AI has been rapidly adopted, its effect has been to shift, not replace the role of the human (see the “decide-execute-deliver” framework normaltech.ai/p/why-ai-hasnt…) Similarly, the panic about what to major in is also misplaced. There will be enduring demand for computer science, philosophy, and just about everything else. (In fact, AI companies hiring philosophers has been a big recent trend.) The bad news AI seems to help senior people much more than juniors. I can use AI for coding because I spent 25 years learning how to code, which lets me supervise coding agents effectively. (See my post on the “growth cycle” vs the “dependence spiral” teapaw.org/random_walker/status/2…) You are in a bind — you can’t offload your skill-building to AI, but you’ll graduate into a market where employers will expect you to get work done with AI. We never faced this dilemma. As a result we haven’t figured out how to revamp our classes to help you do both. You’ll have to help us figure it out. And you’ll need to somehow resist the constant temptation to turn to the shortcut machine. The hope My point is not that AI is bad for learning. It’s an incredibly flexible tool. Is the internet good or bad for learning? Depends — are you using it to find research papers or waste time scrolling? I use AI every day for learning. The key is to use it to increase, not decrease your cognitive load. To learn deeper, not faster. There is no learning without the cognitive sweat. I try to make sure I’m mentally exhausted at the end of the day. I do feel that AI lets me push myself harder than I ever could before, and I have a vision that as AI continues to advance it will enable human-AI “co-superintelligence“. (I talked about this at the end of my ICML keynote. normaltech.ai/p/what-will-be…)
There’s a big, under-appreciated reason why people may have very different experiences and opinions about using AI for work — are they using it for tasks they’re already an expert at, or tasks they can’t do themselves? The former leads to a *growth cycle* and the latter leads to a *dependence spiral*. When I use AI to do something I’m an expert at, like coding, I treat it as a tool. I can build quickly, maintaining an understanding of the code, knowing that if necessary, I can fix the code myself. It feels empowering. It frees up my time to think about the complex, judgment-oriented parts of software engineering that I can’t or won’t delegate to AI. That means my own skills improve rapidly, and I get to climb the ladder of complexity and develop higher-level skills, much more so than when I write the code myself. I feel in control. I can lock in and achieve a flow state — when AI is working, I’m reviewing, building understanding, and planning the next steps. I never get the feeling that the tool is about to replace me. This is the growth cycle. (Of course, the growth cycle is not automatic. I still need to exercise agency to use AI responsibly. But it’s the same challenge with any productivity-enhancing technology, and those who’ve navigated such transitions before are well-equipped to navigate it with AI as well.) On the other hand, if I use it for tasks I don’t understand and haven’t learned to perform myself, I have no choice but to treat it as a superintelligence. If something breaks, the best I can do is ask AI to fix it and hope for the best. I generally can’t evaluate the quality of the output myself. The only way to find out if it's any good is if and when the work is ultimately reviewed by an actual expert. The experience is confusing, unsettling and disempowering. And forget about flow state. By over-relying on AI, I risk losing whatever skill I had at the task in the first place, even if it boosts productivity in the short term. This is the dependence spiral. It’s no wonder that entry-level workers and students preparing to enter the workforce find themselves in a bind. To compete with the AI-enabled productivity of more seasoned workers, they must adopt AI themselves, but doing so risks the dependence spiral. I have some thoughts on solutions that I will share in later posts, but I think having a clear diagnosis of the problem is a useful first step.
73
300
42
1,247
320,496
Linna An retweeted
🚀Please help share widely! We're excited to launch the new #AI4BIO #Fellows Program @SCSatCMU to recruit exceptional early-career scientists pursuing bold, independent research at the intersection of AI and biology. Fellows will be supported by The Center for AI-Driven Biomedical Research (#AI4BIO) @SCSatCMU and co-mentored by two CMU School of Computer Science faculty members, with opportunities spanning AI models, computational biology, and autonomous science, including engagement with the CMU AI Science Foundry (ai-science-foundry.cmu.edu/). We are looking for truly exceptional candidates who want to help define new directions for AI-driven biomedical discovery. 📅 Apply by November 15, 2026. 🧬 Program information: cmu.edu/ai4bio/apply/index.h… 🤖 Application via Interfolio: apply.interfolio.com/190197
7
77
2
259
28,535
One of our most popular preprints of 2025 is now published in @ScienceMagazine. Congrats to @BrianHie, @samuelhking, and team on establishing a framework for generating and evaluating AI-generated genomes. - Access the paper: science.org/doi/10.1126/scie…
In a new preprint from @brianhie’s lab, the team reports the first generative design of viable bacteriophage genomes. Leveraging Evo 1 & Evo 2, they generated whole genome sequences, resulting in 16 viable phages with distinct genomic architectures.
10
44
8
194
65,573
Thrilled to release two new preprints on intelligent labs for driving science and innovation. This is in close coordination with Aviv Regev, Jian Ma (@jmuiuc), Michelle Lee (@michellearning), and the teams at @Genentech, @SCSatCMU, and @Princeton University. In our perspective, we argue that the next generation of labs should be human-in-the-lead and AI-empowered, integrating human intent, machine reasoning, and physical experimentation through scientific world models and an agentic harnessing layer. Done responsibly, these systems have the potential to make scientific discovery more programmable, reproducible, adaptive, and scalable while enabling scientists to focus on higher-level scientific reasoning and discovery. It's been a privilege to pursue this Perspective with @MengdiWang10 and an outstanding group of scientists and innovators advancing the intersection of computation, AI, science, and medicine. We're excited to continue exploring where this vision leads. Our preprint: preprints.org/manuscript/202… Preprint led by Jian and Aviv team: preprints.org/manuscript/202… This also kicks off a new Gladstone-Stanford AI Hub efforts, led by Katie Pollard at @GladstoneInst and myself, with an amazing team of scientists including Emma Lundberg (@Prof_Lundberg), Brian L Trippe (@brianltrippe ), Anshul Kundaje (@anshulkundaje ), Barbara Engelhardt (@BeEngelhardt), Christina Theodoris (@TheodorisLab), Catherine Tcheandjieu (@ines_catherine), Bruce Conklin, Alexander Marson, Stacie Dodgson (@StacieDodgson), Seth Shipman (@seth_shipman), Vijay Ramani, Danielle Swaney (@dlswaney), and Nevan Krogan. Excited to be building together across two great institutions!
1
53
3
224
201,812
Linna An retweeted
It’s getting hard to keep track of all the new open-source cofolding models coming out. To help everyone stay up to date, I'm excited to share cofoldarena.ai, a PDB-synced leaderboard of open models, updated weekly. All predictions are available to view & download! (1/4)
25
99
14
438
51,797
1/ What could we discover if we had a binder for every protein in the human proteome? We are excited to introduce the Bindome, an open and free resource of over 300,000 designed protein binder candidates against over 8,000 human target proteins. 🧵
20
113
8
496
48,725
Big news from Boltz - our biggest update yet! 🚀 Today we’re releasing two new state-of-the-art models for protein and small molecule design with extensive wet lab validation and a new API to run all of our models on scalable GPUs wherever you (or your agents) work! 🔥
36
241
30
1,138
141,095
[ML4BioChem club] 3rd online speaker! @m1nj12, Colin Kalicki!
3
22
1,764
[ML4BioChem club] 2nd speaker!
1
4
35
2,108
First guest for ML4BioChem club comes today! @r_krishna3 ! See you all online and in person.
3
30
2,089
Meet evedesign: a new open-source ML framework that makes protein design accessible and interoperable. 📢 See our post: deboramarkslab.substack.com/… Protein design models are powerful, but combining them shouldn’t require custom glue code. ✅Combine models for multi-objective optimization ✅Integrate lab-in-the-loop experimental of data ✅100% secure: run on your own infra, no data sharing Get started building therapeutics & industrial enzymes today 👇 📄Paper: biorxiv.org/content/10.64898… 💻Code: github.com/evedesignbio 🌐Webserver: evedesign.bio Reach out to collaborate: [email protected]
2
64
2
261
16,578
This is an exciting work! I doubt it will resolve drug discovery, but will be really exciting for enzyme and molecular machine design.
Learning the All-Atom Equilibrium Distribution of Biomolecular Interactions at Scale 1 ByteDance AI Drug Discovery and Anew Therapeutics researchers introduce AnewSampling, the first generative foundation framework that faithfully reproduces molecular dynamics (MD) at the all-atom level for sampling the equilibrium distribution of biomolecular interactions, addressing the high computational cost of traditional MD simulations. 2 AnewSampling leverages a novel quotient-space generative framework to ensure mathematical consistency in its modeling, and is trained on AnewSamplingDB—the largest self-curated database of protein-ligand trajectories to date, containing over 15 million conformations across 10,297 unique protein sequences and 27,979 unique ligand SMILES. 3 The framework builds on an AlphaFold3-like architecture with a stratified hybrid fine-tuning strategy: Low-Rank Adaptation (LoRA) for sequence representation modules and full-parameter fine-tuning for the Diffusion Module, alongside a Cluster-Based Template Guidance mechanism to enforce exhaustive exploration of the equilibrium ensemble. 4 In benchmarking on the ATLAS monomer dataset, AnewSampling outperforms all state-of-the-art generative methods across all 13 evaluation metrics, showing unparalleled accuracy in predicting protein flexibility and distributional accuracy for monomeric systems. 5 For protein-ligand dynamics testing across held-out PDB systems, JACS & Merck industrial datasets and an in-house drug discovery pipeline dataset, AnewSampling achieves statistical alignment with ground-truth MD distributions that far surpasses static predictors and MD-enhanced models like Boltz2, with its generated conformations nearly indistinguishable from MD baselines in key metrics. 6 AnewSampling demonstrates emergent enhanced sampling capabilities beyond conventional MD, successfully navigating high energy barriers to recover coupled ligand and side-chain motions in CDK2 systems (1H1R and 1H1S)—a major challenge for traditional MD that often requires replica-exchange MD (REMD) to achieve. 7 The model accurately captures subtle ligand-induced conformational shifts in congeneric structure-activity relationship (SAR) series, a critical capability for lead optimization in drug discovery, and maintains high fidelity in modeling non-covalent protein-ligand interactions and global protein backbone dynamics across diverse chemical and conformational spaces. 8 The research team proposes a multi-level assessment strategy for generative biomolecular dynamics models, using metrics like Jensen-Shannon (JS) distance for ligand torsion, Wasserstein (WS) distance for protein-ligand interactions and Spearman correlation for Cα RMSF to rigorously validate physical fidelity at the atomic level. 9 AnewSampling offers unprecedented computational efficiency for exploring biomolecular conformational landscapes, enabling integration into research and industrial drug discovery pipelines and driving a shift toward dynamics-aware design of adaptive inhibitors and functional biomolecules. 10 While AnewSampling achieves significant advances, the researchers note current limitations including reliance on structural templates, limited training data for broader biomolecular interaction types (e.g., protein-nucleic acid) and restriction to fixed thermodynamic environments, outlining future work to address these and enable sequence-only equilibrium distribution prediction. 11 AnewSampling and conventional MD are shown to be complementary: MD provides the critical training data for the generative model, while AnewSampling can accelerate MD by generating diverse initial structural candidates that help bypass energy barriers in physical simulations. 📜Paper: biorxiv.org/content/10.64898… #AIDrugDiscovery #BiomolecularDynamics #AllAtomModeling #GenerativeAI #ComputationalBiology #MolecularDynamics #ProteinLigandInteractions
4
42
5,746
Learning the All-Atom Equilibrium Distribution of Biomolecular Interactions at Scale 1 ByteDance AI Drug Discovery and Anew Therapeutics researchers introduce AnewSampling, the first generative foundation framework that faithfully reproduces molecular dynamics (MD) at the all-atom level for sampling the equilibrium distribution of biomolecular interactions, addressing the high computational cost of traditional MD simulations. 2 AnewSampling leverages a novel quotient-space generative framework to ensure mathematical consistency in its modeling, and is trained on AnewSamplingDB—the largest self-curated database of protein-ligand trajectories to date, containing over 15 million conformations across 10,297 unique protein sequences and 27,979 unique ligand SMILES. 3 The framework builds on an AlphaFold3-like architecture with a stratified hybrid fine-tuning strategy: Low-Rank Adaptation (LoRA) for sequence representation modules and full-parameter fine-tuning for the Diffusion Module, alongside a Cluster-Based Template Guidance mechanism to enforce exhaustive exploration of the equilibrium ensemble. 4 In benchmarking on the ATLAS monomer dataset, AnewSampling outperforms all state-of-the-art generative methods across all 13 evaluation metrics, showing unparalleled accuracy in predicting protein flexibility and distributional accuracy for monomeric systems. 5 For protein-ligand dynamics testing across held-out PDB systems, JACS & Merck industrial datasets and an in-house drug discovery pipeline dataset, AnewSampling achieves statistical alignment with ground-truth MD distributions that far surpasses static predictors and MD-enhanced models like Boltz2, with its generated conformations nearly indistinguishable from MD baselines in key metrics. 6 AnewSampling demonstrates emergent enhanced sampling capabilities beyond conventional MD, successfully navigating high energy barriers to recover coupled ligand and side-chain motions in CDK2 systems (1H1R and 1H1S)—a major challenge for traditional MD that often requires replica-exchange MD (REMD) to achieve. 7 The model accurately captures subtle ligand-induced conformational shifts in congeneric structure-activity relationship (SAR) series, a critical capability for lead optimization in drug discovery, and maintains high fidelity in modeling non-covalent protein-ligand interactions and global protein backbone dynamics across diverse chemical and conformational spaces. 8 The research team proposes a multi-level assessment strategy for generative biomolecular dynamics models, using metrics like Jensen-Shannon (JS) distance for ligand torsion, Wasserstein (WS) distance for protein-ligand interactions and Spearman correlation for Cα RMSF to rigorously validate physical fidelity at the atomic level. 9 AnewSampling offers unprecedented computational efficiency for exploring biomolecular conformational landscapes, enabling integration into research and industrial drug discovery pipelines and driving a shift toward dynamics-aware design of adaptive inhibitors and functional biomolecules. 10 While AnewSampling achieves significant advances, the researchers note current limitations including reliance on structural templates, limited training data for broader biomolecular interaction types (e.g., protein-nucleic acid) and restriction to fixed thermodynamic environments, outlining future work to address these and enable sequence-only equilibrium distribution prediction. 11 AnewSampling and conventional MD are shown to be complementary: MD provides the critical training data for the generative model, while AnewSampling can accelerate MD by generating diverse initial structural candidates that help bypass energy barriers in physical simulations. 📜Paper: biorxiv.org/content/10.64898… #AIDrugDiscovery #BiomolecularDynamics #AllAtomModeling #GenerativeAI #ComputationalBiology #MolecularDynamics #ProteinLigandInteractions
3
22
2
81
10,119
I'm excited to announce some major updates to our ProteinEBM paper with Chenxi Ou and @sokrypton!
5
57
1
363
35,451
Linna An retweeted
This is really cool (and wild): Scientists simulated a complete living cell for the first time. Every molecule, every reaction, from DNA replication to cell division. The paper (Luthey-Schulten et al., Cell 2026, doi.org/10.1016/j.cell.2026.…), just out today, used JCVI-Syn3A — a synthetic minimal bacterium with fewer than 500 genes. A 3D+time simulation of the full 105-minute cell cycle: DNA replication, protein translation, metabolism, division. Every gene, protein, RNA, and chemical reaction tracked through physical space. It took years to build. Multiple GPUs. Six days of compute time per run. And this is the simplest possible cell. A human cell has ~20,000 genes. It lives in tissue. It interacts with neighbors. It differentiates. It responds to drugs in ways that depend on context we haven't fully measured. Mechanistic simulation of the minimal cell costs 6 GPU-days for 105 minutes of biology. You cannot scale that to human cells. The complexity isn't 40x harder. It's exponentially harder. This is why the field pivoted to data-driven models. You can't hand-encode the regulatory wiring of a human hepatocyte. But you can learn it — if you have the right perturbation data collected across enough diverse biological contexts. The two approaches aren't competing. Papers like this generate the ground truth that future ML models need for validation. But the path to a clinically useful virtual cell runs through foundation models, not through scaling up mechanistic simulation. Amazing work!
72
511
75
2,101
291,377
We = Linna An (AP@Rice) Cameron Glasscock(AP@Rice) Hayden Stegall(grad@Rice BCB); Precious Castillo (grad@Rice BCB) Yanapat Janthana (grad@Rice SSPB); Welcome sponsors💸; welcome speaker nomination!
We are launching ML4BioChem club to connect scientists through in-person/online seminars and happy hours to spark new collaborations, and beyond. Upcoming speakers 👉 : linnaan-lab.github.io/ML4BIO… Subscribe @ `Email me about events` on website. Will add you to [email protected].
5
989
We are launching ML4BioChem club to connect scientists through in-person/online seminars and happy hours to spark new collaborations, and beyond. Upcoming speakers 👉 : linnaan-lab.github.io/ML4BIO… Subscribe @ `Email me about events` on website. Will add you to [email protected].
3
20
2
161
10,173