Pre-synthesis stability screen
For four questions an assay cannot answer cheaply: whether a fold holds, whether a multimer interface survives, whether a metal site stays coordinated, and whether a construct that will not express is worth another attempt. Candidates go through construct clean-up, 3D structure prediction, and a two-stage explicit-solvent molecular dynamics stress test in water and 150 mM NaCl — held at 300 K, then ramped to 330 K (57 °C) to see what comes apart. You choose how long.
If your question is binding affinity, consider measuring it rather than predicting it. For small, well-behaved binders that express cleanly against a single known target, high-throughput services now return a measured KD at a price per candidate that no prediction can honestly compete with, and published benchmarks show computational binding metrics correlating only weakly with the measured value. We would rather say that here than take the engagement. What those services do not cover is the list above: a construct that will not express, a multi-chain assembly, a metal-loaded site, or a question about whether a fold survives heat rather than whether it binds. That is where simulation earns its place, and it is the only place we will sell it to you.
| Tier | Per candidate | The question it answers | Typical use |
|---|---|---|---|
| Triage | 0.5 ns | Does the fold survive contact with solvent and a thermal ramp at all? | A whole screen, 20–50 candidates |
| Standard | 10 ns | Once the fast failures are gone, does it still hold its shape and its interface? | A shortlist of about ten |
| Extended | 50 ns+ ×3 | Does the answer reproduce across independent runs, or did one trajectory get lucky? | The 3–5 you are about to commit budget to |
Extended runs are triplicate with independent starting velocities, and we report the spread across replicas rather than a single mean, so you can judge how firm the result is. Most engagements run Triage across the full set, then take the survivors up a tier.
You send
- 20–50 sequences in FASTA
- Intended expression host, if known
- Any structures you already have
- Oligomeric state, if the target is a multimer
We measure
- Backbone Cα RMSD against the starting model
- Radius of gyration drift
- Inter-chain contact retention, for multimers
- Metal-to-donor distances, for metal sites
You receive
- Ranked table, best to worst, with the basis for each rank
- One page per candidate
- Every 3D model as PDB
- All metrics as JSON, so you can re-plot them
- A methods section you can paste into a report
Each candidate lands in one of three bands. The thresholds are fixed before your run, not chosen afterwards to suit the results:
Marginal mean Cα RMSD ≤ 2.6 Å at 300 K, contact retention ≥ 75%
Fails everything else — loses its fold or its interface under the ramp
The 3–5 day figure is Triage across a full screen: three business days for up to 24 sequences, five for 25–50, measured on our own hardware rather than estimated. Standard and Extended tiers run longer and we quote their schedule in writing at intake, because it depends on sequence length, oligomeric state and how much GPU capacity you want us to add. Fee scales with tier, sequence count, and whether multimer interfaces or metal centres are involved.
Start this engagement →No tier here measures thermodynamic stability. Triage is a short-timescale filter: it separates folds that come apart quickly under a thermal ramp from folds that hold their shape, which is good at catching candidates that were never going to work and is not evidence that a survivor is stable. Standard and Extended buy more confidence in that ranking, not a different kind of answer. None of it predicts expression yield, solubility in your buffer, aggregation over days, or catalytic activity. Treat the output as a ranking that tells you where to spend synthesis budget first, not as a substitute for an assay.