Your protein.
Testable hypotheses.
The next experiment series.

Start with a protein and goal

Enter a UniProt accession. For primers, add your full CDS or explore a synthetic gene design. Define application conditions and success criteria for the laboratory series.

Find donor candidates

The current beta searches up to 80 reviewed entries with the same EC number, comparing sequences and up to six structures. You can choose a homolog manually. This limited search cannot guarantee the best donor.

Temperature and pH information is shown when available in annotations. Similarity and origin from a thermophile alone do not establish the desired stability.

Compare and screen functional sites

Retrieve existing AlphaFold DB models and UniProt annotations. Align residues, overlay structures and inspect confidence. Active sites, binding sites and other functional regions are excluded from the main list. Switch between cartoon, surface and sticks, overlay a homolog and highlight differences, deviations or confidence. The legend explains each color.

Rank with disagreements visible

The linear Foldiff model scores the residue environment and substitution features. The ThermoMPNN neural network adds a second prediction when available. FireProtDB measurements are shown separately with conditions and sources.

Negative ΔΔG suggests stabilization. It cannot be converted into a gain in degrees or activity. Model agreement is not a success probability either.

Select variants and a control

Select candidates, add the WT protein and replicates. Export the series table and primer designs. Distances help reveal contacts between positions; combination effects require separate testing.

Bring experiment results back

Enter measurements or import CSV. The service shows means, replicate spread and differences from WT. Measure activity separately. For pH, view a curve and the maximum among tested points.

Distinguish measured candidates from hypotheses

Compare the measured effect and retained activity against your threshold. Show untested variants with agreeing predictions separately. Confirm promising variants in a repeat series. There is no automatic model retraining yet.