Molence (@MolenceBio) header
Molence (@MolenceBio) profile picture

Molence

@MolenceBio
molence.com/
Joined October 2026
  • Tweets23
  • Following109
  • Followers676
21Media
Photo by Molence (@MolenceBio)Photo by Molence (@MolenceBio)Photo by Molence (@MolenceBio)Photo by Molence (@MolenceBio)Photo by Molence (@MolenceBio)Photo by Molence (@MolenceBio)Photo by Molence (@MolenceBio)Photo by Molence (@MolenceBio)Photo by Molence (@MolenceBio)
Introducing Mol-1: Open Frontier Biomolecular Intelligence 🚀🚀🚀 📊 75.0% on FoldBench-AB, near IsoDDE's reported 75.58%!!! 🧬 60k-token assemblies on one single H20, outperforming Anthropic 8/10. 🎯 VHHs binding hit rate as high as 63% with best SPR KD at 43 pM. GitHub: github.com/molence/Mol-1 Try it: lab.molence.com
16
99
13
594
63,882
🧬 Mol-1 turns additional inference compute into better antibody–antigen predictions. Two complementary ways to scale: sample more candidates and refine more deeply. On MolBench-AB-26: 📈 More sampling: Increasing from 1 to 100 seeds raises mean DockQ from 0.359 to 0.463, with prediction quality continuing to improve through 100 seeds. 🔄 Deeper refinement: At a fixed 5-seed budget, increasing recycling from 10 → 20 → 30 cycles raises the success rate from 53.5% → 57.3% → 58.7%. ⚡ Better sampling efficiency: 30 cycles × 5 seeds achieves 58.64% success, compared with 58.69% for 10 cycles × 40 seeds—comparable accuracy with 8× fewer seeds. Mol-1 gives researchers two practical ways to turn inference compute into stronger molecular predictions. Code & weights: github.com/molence/Mol-1 #AIforScience #ProteinStructure
Antibody–antigen test-time scaling on MolBench-AB-26. Mean DockQ is plotted against the number of sampled seeds from 1 to 100 on a logarithmic axis. Blue curves show Mol-1 with 10, 20, and 30 recycling cycles; gray curves show OpenDDE and ESMFold2. Default 20-cycle Mol-1 improves from 0.359 mean DockQ at 1 seed to 0.463 at 100 seeds. For each pair, the highest-confidence candidate is selected from K seeds with 5 samples per seed.

ALT Antibody–antigen test-time scaling on MolBench-AB-26. Mean DockQ is plotted against the number of sampled seeds from 1 to 100 on a logarithmic axis. Blue curves show Mol-1 with 10, 20, and 30 recycling cycles; gray curves show OpenDDE and ESMFold2. Default 20-cycle Mol-1 improves from 0.359 mean DockQ at 1 seed to 0.463 at 100 seeds. For each pair, the highest-confidence candidate is selected from K seeds with 5 samples per seed.

Recycling and sampling efficiency of Mol-1 on MolBench-AB-26. Success means DockQ greater than 0.23, with 5 samples per seed. Three configurations achieve similar success rates: 30 cycles × 5 seeds, 58.64%; 20 cycles × 10 seeds, 58.83%; and 10 cycles × 40 seeds, 58.69%. Their horizontal positions correspond to cycle × seed budgets of 150, 200, and 400. Deeper recycling achieves comparable accuracy with fewer seeds.

ALT Recycling and sampling efficiency of Mol-1 on MolBench-AB-26. Success means DockQ greater than 0.23, with 5 samples per seed. Three configurations achieve similar success rates: 30 cycles × 5 seeds, 58.64%; 20 cycles × 10 seeds, 58.83%; and 10 cycles × 40 seeds, 58.69%. Their horizontal positions correspond to cycle × seed budgets of 150, 200, and 400. Deeper recycling achieves comparable accuracy with fewer seeds.

2
4
1
35
1,491
Mol-1 agent: large protein assemblies on a single NVIDIA H20, outperforming Anthropic 9/10. In our 10-case evaluation (8,000-60,120 residues), 8 assemblies reached TM-score ≥0.9. The five 30K-60K-residue capsid cases scored 0.993-0.999; Anthropic's published runs for these cases produced collapsed structures. The ~10K group is mixed: Mol-1 has higher reported TM-scores on 4 of 5 cases, but is lower on 7UTD (0.478 vs 0.96). These results come from a long-running, iterative prediction-and-assembly workflow, taking tens of hours. The figure compares published results under different hardware and inference settings; it is not a controlled head-to-head benchmark. Reports: Mol-1: github.com/molence/Mol-1/blo… Anthropic: anthropic.com/research/claud…
Published large-complex predictions in a 5-column, 4-row grid. Each pair of rows compares Anthropic (above) with Mol-1 agent (below) for the same PDB. Top group, Anthropic/Mol-1 TM-scores: 7NAP 0.69/0.985; 9WIE 0.73/0.981; 7VCU 0.54/0.914; 8SU9 0.81/0.895; 7UTD 0.96/0.478. Bottom group: Anthropic reports collapsed structures for all five cases; Mol-1 scores are 1LP3 0.999, 9C3J 0.997, 6HTX 0.995, 9EQW 0.997, 9P4N 0.993. The largest is 9EQW: 60,120 residues, 120 chains. Anthropic used 8 H100/B200 GPUs for the smaller group and 8 B300 GPUs for the larger group; Mol-1 used one H20. Different inference settings; cross-report comparison. Original views are independently scaled. Mol-1 blue is prediction, gray is experiment. Sources: Anthropic report Figures 7-8; Mol-1 report Figure 6.

ALT Published large-complex predictions in a 5-column, 4-row grid. Each pair of rows compares Anthropic (above) with Mol-1 agent (below) for the same PDB. Top group, Anthropic/Mol-1 TM-scores: 7NAP 0.69/0.985; 9WIE 0.73/0.981; 7VCU 0.54/0.914; 8SU9 0.81/0.895; 7UTD 0.96/0.478. Bottom group: Anthropic reports collapsed structures for all five cases; Mol-1 scores are 1LP3 0.999, 9C3J 0.997, 6HTX 0.995, 9EQW 0.997, 9P4N 0.993. The largest is 9EQW: 60,120 residues, 120 chains. Anthropic used 8 H100/B200 GPUs for the smaller group and 8 B300 GPUs for the larger group; Mol-1 used one H20. Different inference settings; cross-report comparison. Original views are independently scaled. Mol-1 blue is prediction, gray is experiment. Sources: Anthropic report Figures 7-8; Mol-1 report Figure 6.

1
14
137
7,563
1/15 🧬 Mol-1 on antibody structure prediction: on par with IsoDDE, open to build on. 75.0% median success at DockQ > 0.23, on par with IsoDDE’s reported 75.58%. Across all 201,376 five-seed combinations from 32 seeds, 62.22% matched or exceeded 75% success—showing robust performance across seed selections. We release the model code and weights so researchers can run Mol-1 in their own environment and build on its capabilities. Code & weights: github.com/molence/Mol-1 Try Mol-1: lab.molence.com #AIforScience #DrugDiscovery
Two panels show FoldBench-AB success rates and the distribution over all 201,376 five-seed combinations from 32 seeds. Mol-1 scores 75.0%, compared with IsoDDE's plotted 75.6%; Mol-1 with constraints scores 88.3%. In the distribution, 62.22% of combinations reach at least 75% success and 17.85% reach at least 76%.

ALT Two panels show FoldBench-AB success rates and the distribution over all 201,376 five-seed combinations from 32 seeds. Mol-1 scores 75.0%, compared with IsoDDE's plotted 75.6%; Mol-1 with constraints scores 88.3%. In the distribution, 62.22% of combinations reach at least 75% success and 17.85% reach at least 76%.

3
8
6
46
7,092
more replies
14/15 📍 Bring epitope knowledge into structure prediction. With epitope information, Mol-1-constraint improves antibody–antigen success at DockQ > 0.23 across all four benchmarks: • FoldBench-AB: 75.0% → 88.3% • AF3-AB: 62.7% → 84.2% • PXMeter-AB: 58.3% → 77.5% • MolBench-AB-26: 57.3% → 79.4% Evaluate Mol-1 and Mol-1-constraint with your own biological questions: lab.molence.com #AIforScience
Four antibody-antigen benchmark charts show success rates split by DockQ accuracy level. Adding epitope constraints increases Mol-1 success from 75.0% to 88.3% on FoldBench-AB, 62.7% to 84.2% on AF3-AB, 58.3% to 77.5% on PXMeter-AB and 57.3% to 79.4% on MolBench-AB-26.

ALT Four antibody-antigen benchmark charts show success rates split by DockQ accuracy level. Adding epitope constraints increases Mol-1 success from 75.0% to 88.3% on FoldBench-AB, 62.7% to 84.2% on AF3-AB, 58.3% to 77.5% on PXMeter-AB and 57.3% to 79.4% on MolBench-AB-26.

1
2
165
15/15 📍 Three residues can provide a useful structural clue. In three case studies, we supplied three-residue site hypotheses informed by literature and monomer structures. With Mol-1-constraint, DockQ improved: • 0.014 → 0.874 • 0.011 → 0.739 • 0.200 → 0.863 These examples show how sparse biological hypotheses can guide complex prediction. #AIforScience
Three structural case studies compare Mol-1, sparse epitope guidance and oracle epitope guidance. Sparse three-residue site hypotheses improve DockQ from 0.014 to 0.874 for 7UA2, from 0.011 to 0.739 for 9Z76 and from 0.200 to 0.863 for 9GEI. Predicted structures are overlaid on gray references.

ALT Three structural case studies compare Mol-1, sparse epitope guidance and oracle epitope guidance. Sparse three-residue site hypotheses improve DockQ from 0.014 to 0.874 for 7UA2, from 0.011 to 0.739 for 9Z76 and from 0.200 to 0.863 for 9GEI. Predicted structures are overlaid on gray references.

1
155
Load more