Protein Machines

Prioritizing natural proteins
for experimental screening

From protein sequences to a shortlist you can test in your own lab.

Discuss your screening decision

No confidential data needed for an initial conversation.

Workflow: an unranked panel of natural protein sequences found in nature, together with initial assay measurements (sequences plus measured activity or solubility), are inputs — no candidate is favoured before ranking. The ranking stage predicts assay performance and reorders the same candidates; rankings are predictions, not confirmed results. A shortlist of the highest-ranked candidates is selected within your testing budget — the shortlist is the framework's output. Experimental validation then measures the chosen property in your lab, and new results can inform the next round.

We prioritize existing natural proteins for testing.

Does ranking improve the shortlist?

Compare methods within each row; tasks use different measurements.

Esterase screening

Active hits per 10 selections

Picking by chance
3.57
Sequence-feature model
6.86
AI model
5.74

eSOL solubility

Mean solubility (%)

Picking by chance
50.1
Sequence-feature model
79.2
AI model
82.9

ADK activity

Turnover summary (s⁻¹)

Picking by chance
73.8
Sequence-feature model
95.2
AI model
106.1

Picking by chance selects ten candidates without a ranking model.

The next question: does this improve your screening workflow?

Discuss your screening decision

No confidential data needed for an initial conversation.