←
Geodesic Intelligence19 posts
Geodesic Intelligence avatar

Geodesic Intelligence

@Geodesiclab

Build AGI that discovers the shortest path from biology to medicines

◷ Joined July 2026▧ 8 Photos and videos
Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Nov 10

Introducing NovaAtom-Lite, an all-atom structure prediction model for drug discovery. It predicts the structures of proteins, nucleic acids, and the molecules they bind, including how they fit together. NovaAtom-Lite outperforms both IsoDDE and AlphaFold 3 on FoldBench antibody–antigen structure prediction (DockQ ≥ 0.23): NovaAtom-Lite: 79.07% (n=172) IsoDDE: 75.58% AlphaFold 3: 47.90% (n=167) Our evaluation uses 5 seeds × 5 samples, with the top-1 prediction selected by the model's own ranking score. Better structure prediction. A stronger starting point for your next experiment. Live today via Geodesic API and NovaDDE. 🔗 API: platform.geodesiclab.com/ 🔗 NovaDDE: agent.geodesiclab.com 🔗 Model Card: geodesiclab.com/news/introdu…

Post attachment
Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Nov 16

The #BeyondBinding Challenge is officially underway! Excited to partner with @GenScript to put AI-designed antibodies to the test. 🎯 Target: IL-4Rα ECD 🧪 Submit up to 20 AI-designed IgG antibodies for experimental testing 🏆 Top teams advance to Round 2 Round 1 deadline: Nov 23, 11:59 PM ET. Looking forward to seeing the community's designs and experimental results!

The @GenScript x @Geodesiclab #BeyondBinding Challenge is officially on 🧬 Target: IL-4Rα ECD. Submit up to 20 AI-designed IgG antibodies. We’ll test selected designs in the lab. Top teams advance to Round 2. 📅 Round 1 deadline: Nov 23, 11:59 PM ET 👉 bit.ly/4iQTjxw

Post attachmentPost attachmentPost attachment
GenScript avatar
⟳ Geodesic Intelligence retweeted
GenScript@GenScript·Nov 11

⏰ Last chance! Registration closes tomorrow! Can your AI design antibodies beyond binding? 🧬 Binding + IgG developability 🔬 Wet-lab data to refine AI models 🤖 AI credits + $2.5M in awards & support Join @GenScript × @Geodesiclab! Apply: bit.ly/4zCje2P

Post attachment
Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Nov 10

Runs N’ Poses tests how well models predict protein–ligand structures on examples that look different from the training data. The chart goes from the least familiar cases on the left to the closest matches on the right. NovaAtom-Lite leads across all eight groups: • Least familiar (n=63): NovaAtom-Lite 47.6% vs AlphaFold 3 23.8% · OpenDDE 15.9%—twice AlphaFold 3’s success rate. • Most familiar (n=664): NovaAtom-Lite 91.0% vs AlphaFold 3 88.2%. From the most familiar cases to the least familiar, NovaAtom-Lite’s success rate drops 43 points. The other models drop 64–75 points. It holds up better on examples that look less like the training data. We used the publicly available predictions from AlphaFold 3, Chai-1, and Boltz-1, while running NovaAtom-Lite and OpenDDE ourselves. All results use the same ligands and success criteria: symmetry-corrected RMSD < 2 Å and LDDT-PLI > 0.8.

Post attachment
Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Nov 10

We also tested NovaAtom-Lite on PXMeter’s PXM-22to25-AB antibody–antigen set—528 PDB entries, compared with 172 targets in FoldBench. • NovaAtom-Lite 66.05% (528 entries / 383 clusters) vs OpenDDE 51.0% (494 entries)—a 15-point margin. • ESMFold2 follows at 42.9%, Protenix-v1 at 40.2%, and Boltz-1 at 19.6%. PXMeter is an open-source toolkit for evaluating predicted structures. This set draws on PDB releases from 2022–2025, filtered for low homology to earlier structures. Success here means DockQ ≥ 0.23.

Post attachment
Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Nov 10

Test-time Scaling: NovaAtom-Lite gets better with more seeds. We tested 1 to 100 seeds on the same 172 FoldBench antibody–antigen interfaces throughout. • Top-1 success rises from 74.3% with 1 seed to 83.1% with 100 seeds—an 8.8-point gain (DockQ ≥ 0.23). • Stricter thresholds show the same trend: +8.4 points at DockQ ≥ 0.49 and +5.5 points at DockQ ≥ 0.80. • Going from 1 to 5 seeds already adds 4.0 points—nearly half the total gain—with further improvements through 100 seeds. At 100 seeds, picking the best generated candidate in hindsight gives 88.4% success (oracle), 5.3 points above the model’s top-ranked results. Good predictions are already in the pool; the ranking score just doesn’t always put them first. More seeds give users a way to trade extra compute for better predictions. Better ranking could help us get more from the predictions we already generate.

Post attachment
Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Nov 10

NovaAtom-Lite outperforms competing models on FoldBench across two key drug-discovery tasks: antibody–antigen and protein–ligand structure prediction. • Antibody–antigen: 79.07% (n=172) vs IsoDDE 75.58% · AlphaFold 3 47.90% (n=167). • Protein–ligand: 77.06% (n=558) vs IsoDDE 75.99% · AlphaFold 3 64.90% (n=547). FoldBench covers 1,522 biological assemblies from PDB entries deposited in 2023–2024, filtered for low homology to AlphaFold 3’s training data. Success means DockQ ≥ 0.23 for antibody–antigen prediction, and ligand RMSD < 2 Å plus LDDT-PLI > 0.8 for protein–ligand prediction. Monomers are scored by LDDT. Full eight-task comparison below.

Post attachment
Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Nov 10

Why antibody–antigen? Antibodies are a major class of therapeutics. Knowing where and how they bind their targets helps guide their design. Getting that interface right is one of the toughest challenges in structure prediction: the binding loops (CDRs) are flexible, and there’s little co-evolutionary signal to work with. Here’s NovaAtom-Lite’s prediction for 8IDO (in color), overlaid on the experimental structure (gray). DockQ: 0.839.

0:08
Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Oct 22

Excited to partner with @GenScript to launch the Beyond Binding Challenge! 🧬 AI-designed antibodies need to go beyond binding affinity to become real therapeutics. Let's push the boundaries of AI-driven antibody design and turn computational breakthroughs into real-world impact. Up to $2.5M in awards. Stay tuned!

Binding is just the beginning 🧬 @GenScript and @GeodesicLab are launching the Beyond Binding Challenge to benchmark AI-designed antibodies for binding and developability. Compete for up to $2.5M in awards. Stay tuned: bit.ly/4Bhkvhv #AIProteinDesign #DrugDiscovery

Post attachment
Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Sep 25

What does an AI-native workflow for protein binder design look like in practice? In our latest post, we walk through a pilot design run using NovaDDE: Given a target, the AI agent analyzes the design problem, helps select an epitope and design approach, submits the job, and guides the computational evaluation of the resulting candidates. We also share some of the directions we’re exploring as we continue to develop the workflow. A closer look at NovaDDE and where we’re taking it next: geodesiclab.com/news/novadde…

Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Sep 22

Today, we’re launching the Geodesic API Platform. geodesiclab.com/api For the first time, Geodesic is making our AI models and drug discovery capabilities available through APIs. At its core is NovaAtom, our in-house model for predicting the structures of proteins, small molecules, nucleic acids, and their interactions. Now available in Lite Preview, this is the first time NovaAtom has been made available beyond Geodesic’s own discovery platform. Alongside NovaAtom, we’ve integrated the field’s core models — carefully curated and continuously updated. To use the API: curl -X POST platform.geodesiclab.com/api… \ -H "Authorization: Bearer $GEODESIC_API_KEY" \ -F "model=novaatom" \ -F 'params={"jobName":"rbd-nanobody","sequences":">rbd\nTNLCPFGEVFNATRFASV...\n>nb\nQVQLVESGGGLVQAGGSLR..."}' Poll the job, pull the results, and feed them directly into the next model. Your AI assistant can use the platform directly, too. Connect Claude Code, Codex, or your own agent with one URL and an API key. We manage the integrations and handoffs. You focus on the science. Understand biology. Predict structures. Design proteins. Discover molecules. One key. One interface. Start building today!

Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Sep 17

NovaDDE Plus is now live! Built for scientists who use AI for drug discovery every day — more compute, priority access, concurrent runs, API access, and end-to-end campaign execution. $30/month and sign up by September 25 to get your first month free. Our goal is simple: put the most powerful AI for drug discovery into the hands of every scientist. geodesiclab.com/pricing

Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Sep 16

We’re launching the Geodesic Grand Challenges. What are the hardest questions in drug discovery that, if solved, could change what’s possible over the next 5–10 years? We’re asking you to help define them. Selected questions will become official Geodesic Grand Challenges. Each selected contributor receives 10,000 Geodesic credits. Once the challenges are selected, we’ll open a call for solutions, with a $1 million prize for a verified solution to each challenge. The next breakthrough starts with the right question. Submit yours! geodesiclab.com/grand-challe…

Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Sep 9

5/ But computation alone isn’t enough. Biology has to work in the real world. That’s why we’re building our own wet-lab capabilities alongside our AI stack, and creating a closed learning loop between AI and the physical world: AI → Design → Experiment → Learn → Repeat. Better models generate better experiments. Better experiments generate better data. Better data builds better models.

Geodesic Intelligence avatar
Geodesic Intelligence@Geodesiclab·Sep 9

4/ And the intelligence behind drug discovery shouldn’t come only from general-purpose AI. Today, we’re also releasing NovaAtom-Lite-Preview, the first model in our NovaAtom family of biological foundation models. NovaAtom-Lite-Preview is our starting point, which is built to move fast, learn fast, and give the community an early look at the NovaAtom architecture and capabilities. Larger and substantially more capable NovaAtom models are already on the way.

0:23