AI + experimental drug discovery

Learning how chemistry changes cells.

We make millions of new molecules, test them on living cells, and train AI models on what happens.

measured predicted confirmed in the lab Illustration. Pick a dot to find others that respond the same way, or .
10M+molecules made
8M+experiments run
160K+drug-response profiles made public

The problem

There’s no AlphaFold yet for how drugs act on cells.

AlphaFold worked because fifty years of solved protein structures were there to learn from. Nobody has that kind of data for what drugs do to cells.

1 axis · sequence → structure
Largely solved

Protein structure

~105 structures

AlphaFold was trained on about 170,000 structures that took scientists 50 years to collect.

2 axes · genes × cells
In progress

Cell state

~109 data points

Virtual-cell models are now training on large atlases of gene knockouts.

3 axes · chemistry × targets × context
Open

Cellular pharmacology

chemistry × targets × cell context

Every drug, target and cell type combine differently. A gene knockout is not a drug, and public data covers very little of this.

So we’re generating it.

We built a lab platform that makes large chemical libraries and measures what each molecule does to living cells, much faster and more cheaply than standard screening. It can also read out the molecular changes behind each effect, in the same cells. We design every library so a model can learn from it.

How it works

Each round of experiments decides the next.

We make molecules, test them on cells, train models on the results, and use the models to choose the next batch. Then we go around again.

Step 1 of 5 · Synthesize

Make molecules a model can learn from

We build large libraries of new small molecules, peptides and macrocycles, laid out so we can tell which part of a molecule caused which effect.

What comes out

The same experiments turn up drug leads.

Every screen that trains our models also produces starting points for new drugs, each with structure–activity data and a working idea of how it acts.

Many leads from one screen

A single screen turns up many chemical series, and the structure–activity, mechanism and off-target signals all come from that same experiment.

Predictions before we synthesize

Our models predict what a molecule will do to cells before we make it. That’s how we decide what to make next.

Our own programs

We’re working on targets that have resisted drug discovery for years, and our lead programs are moving into animal studies. We also partner on programs that fit.

Different chemistry, same biology

illustration
Molecule A
structure
Molecule B
structure

Effect on
the cell
Structural overlaplow
Response similarityhigh
Two molecules can look nothing alike and still do the same thing to a cell. Our models find pairs like this, including molecules that mimic a gene knockout or a known drug. Each pair is a hypothesis we can test.

Team

The people behind Zafrens.

Portrait of Swamy Vijayan

Swamy Vijayan

CEO
Portrait of Yi Zhang

Yi Zhang

Chief Platform Officer
Portrait of Joey Azofeifa

Joey Azofeifa

CTO
Portrait of Kevin Eastwood

Kevin Eastwood

CBO
Portrait of Martin Smrcina

Martin Smrcina

Head of Chemistry
Portrait of Maina Ndungu

Maina Ndungu

Head of Library Chemistry
Portrait of Nathan Hedrick

Nathan Hedrick

Head of Screening
Portrait of Paul Laikind

Paul Laikind

Chairman

News

Latest from Zafrens.

For updates, follow Zafrens on LinkedIn.

Contact

Work with us.

We work with pharma, biotech and AI companies on discovery programs, data and licensing. Tell us what you’re working on.

Email hello@zafrens.com
Office 5890 Pacific Center Blvd., Suite 200
San Diego, CA 92121