Mandrake Bio
Editors built for elite plant tissue
Crop context and delivery shape the design from the beginning, rather than being imposed on a finished editor.
We don't search for proteins. We design them.
Our full-stack platform runs from metagenome to validated editor: a richly connected DNA and protein database, function-aware generative models, and an in-house wet lab that closes the loop.
Nature shows us what worked.
We design what should.
Function built into the design
Rather than generating proteins randomly, we encode each application's biological and biophysical requirements directly into our models, then validate function through purpose-built assays. Every result feeds the next design round. This approach is already producing double-digit experimental hit rates.
Nature optimised proteins for survival. We optimise them for purpose.
That is a fundamentally harder problem, and the one we built Mandrake to solve.
Designed to move from possibility to function
Compact editors built around delivery
Size, targeting and cell context become inputs to the design brief, not filters applied at the end.
New systems from biological dark matter
TerraBase makes rare architectures searchable so promising systems can move into design and assay.
Research
Visit the blog
ORBIT · RBX1The only RBX1 design rated Strong
In the GEM × Adaptyv RBX1 competition, Mandrake’s ORBIT produced the only submission Adaptyv classified as Strong.
Read the result
Protein Language Models: Fluent, but clueless
Four experiments on what protein language models learn, where they fail and why structure and physics still matter.
Let's make life programmable
Backed by
