What researchers are trying
TerraBind uses a coarse pocket representation, diffusion-free pose optimization, learned affinity likelihood and calibrated uncertainty for continual learning and hedged batch selection.
Fast, uncertainty-aware affinity screening
Can virtual screening retain useful structural and affinity accuracy without expensive all-atom diffusion for every compound?
Why it matters
Large chemical libraries require a practical speed–accuracy frontier, and uncertainty is needed to decide which predictions deserve synthesis or expensive physics.
Evidence boundary
The results are preprint claims; proprietary assay details and cross-lab prospective replication remain limited.
Organizations represented
Terray Therapeutics research team
Demonstrated evidence
What has actually been shown.
- Preprint-reported 26-fold faster inference and calibrated affinity uncertainty on public and proprietary evaluations.
Unresolved questions
- Does coarse representation preserve water, ion, metal and covalent chemistry?
- How well does uncertainty calibrate under target and chemistry shift?
- Can active learning improve real design-make-test cycles?
Signals to watch next
- Prospective screening campaigns
- Calibration under OOD shift
- Throughput benchmarks
- Active-learning experiments
Connected evidence graph
Related BioAtlas model passports.
Boltz-1 / Boltz-2
Open-source AF3-quality structure — plus binding affinity.
4/7 evidence fields documentedDiffDock
Reframing molecular docking as a diffusion generative problem.
4/7 evidence fields documentedNVIDIA BioNeMo
The GPU-accelerated toolkit that ships biology's foundation models.
2/7 evidence fields documentedNumerion AI Chemistry Platform
The current COSMOS, APEX and EXPO chemistry stack from the company formerly known as Atomwise.
2/7 evidence fields documentedPrimary and evaluation sources