Closed-Loop Protein Discovery

Every experiment
retrains the model.

A generative diffusion model produces binder candidates. A proprietary functional display platform validates them against live cells. Validated hits advance to modular PK scaffolds and in vivo PK/PD studies. Every data point — cell, conjugate, animal — feeds back into the next design cycle.

Generative DesignCell ValidationPK ConjugationIn Vivo PK/PDModel Retraining

The Feedback Problem

Generative models produce candidates faster than any lab can test them. The bottleneck is validation throughput and the absence of a structured feedback signal. Fabricagen is the feedback signal.

01

Cell-level binding data

Per-candidate functional labels from the Cells-on-Array platform. Binding signal per sequence, per receptor, at scale. Not a bulk selection — individual functional measurements.

02

Conjugate characterization

Validated hits are conjugated to modular PK scaffolds. Conjugate yield, homogeneity, and stability data inform which sequence features translate to well-behaved drug candidates.

03

In vivo PK/PD profiles

Conjugates enter rodent PK/PD studies. Exposure, half-life, and pharmacodynamic response data close the loop — the model learns what drives in vivo performance, not just cell binding.

All three data streams retrain the generative model. The platform accumulates target-specific knowledge with every campaign.

Platform

Six components. One loop.

01

Generative Design

A proprietary generative stack designs binder candidates conditioned on antigen and proprietary developability levers. New sequences are produced end-to-end in a single forward pass — not a ranking of known ones.

02

Cells-on-Array

Proprietary functional display platform. Each candidate gets an independent functional binding label against live receptor-expressing cells. Per-sequence measurements, not population averages.

03

In Silico Pre-screen

Generative outputs are ranked by an internal scoring layer before any physical experiment. Assay capacity is spent on the most promising sequences.

04

Developability by Construction

Multi-axis biophysical constraints — charge, pI, cysteine count, germline identity, CDR3 length, chemical liability sites — embedded in the generative prior. Developable candidates are sampled from the model, not selected from it.

05

Modular PK Scaffold Conjugation

Peptide hits are conjugated to modular PK scaffolds using site-specific chemistry. Pharmacokinetics are tuned independently of the binding domain. Antibody-format hits route directly to in vivo studies.

06

In Vivo PK/PD & Loop Closure

Peptide-scaffold conjugates and antibody-format hits enter rodent PK/PD studies. Exposure, half-life, and pharmacodynamic endpoints feed back into model retraining alongside cell assay data.

0 week
Design-to-Cell-Data
Design cycle from generation to functional labels
0
Developability Axes
Conditioned at generation, not post-hoc
0
Binder Formats
VHH, scFv, and Fab from a single model
0 stages
Feedback into Model
Cell data, conjugate data, in vivo PK/PD

Contact

Fabricagen is building infrastructure for closed-loop drug discovery.

If you are a scientist, investor, or potential research partner, we would like to hear from you.

Get in Touch