NB-IRON-07
FORTIFICATIONIron-dense protein bar
Heme-mimic fold that chelates Fe²⁺ from leafy inputs. Animal-source bioavailability, zero animal.
Null Bites is a biocomputation company building nutrition products. We write novel food-grade proteins from first principles, then turn them into things you can eat.
Nutrition is protein engineering. Meat texture comes from myosin and collagen. Eggs set because of ovalbumin. Cheese stretches because of casein. Every property you care about in food traces back to the shape and behaviour of a specific protein.
The tools that design drug molecules can design food molecules. The infrastructure to do it at scale exists because we built it. ProteinMPNN and ESMFold, stitched into an autonomous loop that reads papers, generates sequences, runs simulations, scores every candidate against real constraints, and iterates on its own.
Our designs don't stop at a .pdb file. Each top candidate goes to the bench, into a fermenter, and out the other side as an ingredient formulated into a real product. Performance, recovery, fortification, gut health, plant-based whole foods. The benchmark is simple: you pick it up, you bite into it.
Ignored for
5 years
Since AlphaFold 2, every computational biology dollar chased therapeutics.
Agriculture uses
70%
…of global freshwater. And still can’t scale to 10B people.
It runs experiments, scores its own results, and only the designs a human would trust make it to the bench. Everything that follows ends in a thing you can eat.
An autonomous scientist reads every new paper, writes new amino-acid sequences, and never sleeps.
Every candidate folded in silico and stress-tested. A self-critic throws out 99% as too optimistic.
The survivors are ordered as DNA, expressed in bacteria, purified, and handed to a human food scientist.
Formulated, sensory-panelled, shipped. The only benchmark that counts is a person taking a bite.
1,400+ papers parsed into a living corpus of functional motifs, substrate bindings, and known dead-ends. The reader never sleeps and never forgets a result.
Novel amino-acid sequences generated against a target fold with a bundle of functional constraints: heat stability, water solubility, fibre alignment, binding pocket geometry.
Every sequence is folded in silico, then scored on stability, manufacturability, and nutritional fit. The pipeline throws out anything without a clear physical basis.
146 cycles of self-correction. The system reads its own output and drops candidates that look too confident. Designs that survive cycle 140+ get to the bench.
Top candidates are ordered as DNA from a synthesis vendor, expressed in bacteria, purified, and handed to the formulation team.
The ingredient becomes food. Sensory panels with real people. If it passes, it ships. If it fails, the result goes back to the top of the loop.
Each candidate is a novel sequence, folded in silico, synthesized at the bench, and formulated into a product with a specific job.
Iron-dense protein bar
Heme-mimic fold that chelates Fe²⁺ from leafy inputs. Animal-source bioavailability, zero animal.
Fibrous whole-muscle analogue
Long coiled-coil engineered to self-assemble into aligned fibres under shear. The missing chew.
Heat-set bake gel
Ovalbumin-inspired scaffold tuned for lower set temps. Gels like egg, without touching a hen.
+ 2 more products in queue · Performance · recovery · gut-health · plant-based whole food
Every sequence, every score, every failed synthesis. Published within 48 hours. Forkable. Falsifiable. The product margins are ours. The science isn't.
Pharma kept its wins private and its failures private too. Food is too important for that. Open the science, ship the product, let anyone build the next one.
NB-IRON-07: dock score improved −11.3 → −12.1
pipeline: self-critic caught over-confidence on fold #146
NB-FIBRE-03: simulation @ shear rate 500s⁻¹ — stable
ingest: 17 new papers from NeurIPS bio-track
NB-GEL-02: Tm reduced 72 → 65°C after redesign
bench: gene synthesis order placed for NB-IRON-07
dataset: OpenBio-Nutrition v0.3 pushed