Enzyme stabilization: choosing the first substitution series

Define a measurable outcome first

«Make the enzyme better” is too broad. Choose an endpoint: melting temperature, residual activity after treatment or another predefined parameter. Record temperature, pH, duration, substrate and method. Success in one assay may not transfer to production conditions.

Distinguish stability from activity. A variant with a higher melting temperature may be a less useful catalyst. WT and variant conditions must be comparable. Foldiff records these two measurements in separate columns.

A homolog supplies differences, not proof

For a first trial, enter public UniProt accessions. The current example compares subtilisin P00782 with thermitase P04072. The service compares sequences and existing structures, checking confidence and annotations. Similarity alone does not establish that a substitution will retain function.

The beta donor search covers a limited set of reviewed entries with shared function and a few structures. Thermophilic origin is a useful search direction, but cannot replace stability evidence for a specific construct. Check length, domains and mature-protein numbering.

Read the evidence and disagreements

The linear Foldiff model ranks allowed differences. When its calculation completes, ThermoMPNN supplies a separate stability-change prediction. ThermoMPNN — a research neural network for point substitutions described by its authors in 2024. Predictor agreement is a selection filter, not a validated probability of success.

The two predictions disagree in direction for S125D in the saved example. This cannot support a promise of improvement. Open Evidence, inspect functional sites and reported measurements and conditions, then decide which hypotheses warrant testing.

Build a small, testable series

1. Select the single substitutions you want to study.
2. Create an experiment series: the service adds WT and three replicate fields per sample.
3. Distinguish independent experiments from repeated readings of one sample. The current table does not make that distinction automatically.
4. Review primer designs and their correspondence to your CDS before laboratory work.
5. Return measurements, save CSV and JSON, and repeat-test promising variants separately.

Actual combination effects cannot be obtained by summing single predictions. Current combination designs require their own testing. Likewise, means and standard deviations do not replace significance analysis.

Try it before arranging a pilot

Open the self-guided trial. Explore a worked example, choose two candidates, create a plan and inspect illustrative measurements. Calculating your own public protein requires a suitable homolog and an available structure.

Assess a pilot against an agreed outcome: whether you obtained a reproducible hypothesis list and measured the variants. Foldiff has no validated universal accuracy percentage or completed paid laboratory account yet.