Amiti
Amiti is our molecular embedding model for discovery teams. It supports small molecules, peptides, and antibodies in one workflow, and it includes drug-likeness signals early in screening.
What Is Amiti?
Amiti is the internal model behind AgentCures. It maps proteins, ligands, and assay context into a shared representation so we can compare candidates consistently across projects.
Instead of maintaining separate models for each task, we use one core model and adapt the same embedding space to ranking, filtering, and prioritization. That makes rankings easier to compare and share between teams.
Why Amiti Exists
- Use one model for small molecules, peptides, and antibodies
- Keep computing costs manageable for large screens
- Check drug-like properties early in screening
- Keep a consistent ranking signal across the AgentCures pipeline
Billion-Scale Screening, Made Practical
Traditional virtual screening gives you a list of binders. That is useful, but it is not enough. Many binders fail later because of exposure, safety, or metabolism. Amiti is designed to rank for both binding and developability from the start.
Screens Billions of Molecules
Amiti can run very large batches quickly, so teams can explore broad chemical space without enterprise-level budgets.
ADMET-Aware From the Start
Amiti factors ADMET signals into screening so you can avoid spending time on compounds that look strong early but fail basic drug-likeness checks.
Rank Candidates for Further Testing
Amiti scores potency alongside properties that affect drug development, helping teams choose candidates for further testing.
Amiti-3 vs GNINA
We focus on EF@1%, or enrichment factor in the top 1% of ranked candidates. A score of 1x is equivalent to random selection. Higher is better, because discovery teams spend most of their time on the very top of the list. In the comparisons below, Amiti-3 scores higher than GNINA on both screening benchmarks. It also ranks candidates using ADMET properties.
One Model for Several Types of Treatment
Many drug discovery models are built for only one modality. Amiti uses one shared representation so teams can work across small molecules, peptides, and antibodies without switching tools.
This means Amiti can screen small molecules against a protein target one moment, then evaluate peptide binders or score antibody candidates the next, all within the same framework.
Scientists can use the same workflow to compare different types of candidates.
Supported Modalities
- Small molecules: large-scale virtual screening
- Peptides: cyclic, linear, and stapled binders
- Antibodies: full-length and fragment scoring
- Protein-protein interactions: affinity-oriented ranking
What Sets Amiti Apart
General-Purpose Embeddings
Amiti is not limited to one benchmark task. It provides a consistent representation across targets and modalities.
Efficient at Scale
The encoder architecture supports very large screens while keeping runtime and cost manageable for applied teams.
Unified Screening
Screen small molecules, peptides, and antibodies with the same model.
ADMET Included Early
Drug-likeness is considered during screening so teams can focus on candidates that are more likely to progress.
Built for AgentCures
Amiti powers target ranking, virtual screening, and lead prioritization in the AgentCures workflow.
Built for Real Programs
The model is tuned for day-to-day discovery work, where teams need useful rankings and enough throughput to screen large libraries.
Use Amiti in Your Workflow
Amiti is the model behind the AgentCures platform. If you want to evaluate it for your discovery workflow, get in touch.