The Company Teaching AI to Design a Better Enzyme

Enzymes are the quiet machinery of modern medicine. They stitch together drug molecules, copy DNA in diagnostic tests, and cut costs in factories most people never think about. When an enzyme is a little too slow, or a little too fragile, or works only under expensive conditions, the bill lands somewhere - usually in the price of the finished product. Pando Bioscience, a Watertown, Massachusetts company that went through Y Combinator's Winter 2023 batch, is built around the belief that most of those enzymes are quietly leaving money on the table.

The company's answer combines two things that biotech often keeps in separate rooms: a generative-AI model that proposes new enzyme designs, and a physical screening platform that tests them at high volume. The AI suggests specific mutations meant to improve several properties at once. The screen checks whether those guesses actually work. What comes back feeds the model again.

The problem is throughput, not imagination

Designing an enzyme has never been the hard part. Testing enough of them is. Traditional directed evolution and screening are slow and expensive, which means an AI model trying to learn enzyme behavior is usually starved for data. Pando's central bet is that if you dramatically raise how many variants you can test, the AI finally has enough to learn from - and each cycle makes the next one smarter.

Pando says its ultra-high-throughput "one-pot" screening platform tests roughly 1,000 times more enzyme variants than conventional approaches, about 75% faster and 80% cheaper. The platform's trick is converting many different biological signals into abundant, quantifiable data across varied conditions - the kind of clean numbers a model can actually use.

Why the screen is the moat

Plenty of startups claim "AI for proteins." Fewer talk about the boring machine that feeds the AI. Pando's differentiation sits in that unfashionable place - the screening throughput - because it is what turns a clever model into one that keeps improving. The chart below sketches the company's own framing of that gap.

Two ways to make money

Pando works both sides of the enzyme market. For pharmaceutical, diagnostic, and biotech customers, it improves existing enzymes or discovers new ones - a services and engineering relationship aimed at cutting manufacturing costs and raising productivity. At the same time, it develops its own proprietary enzyme assets. One named example is Saiyan-Phi29, a high-fidelity Phi29 polymerase of the sort used in DNA amplification and sequencing.

That dual model - engineer for clients, and build a catalog you own - is a common shape in the enzyme business, and it hedges a young company between near-term revenue and longer-term product value.

From a synthetic-biology unicorn to a five-person team

Pando was founded in 2022 by Will (Yangxiaolu) Cao and Yang Wang, who met the field the hard way. Cao, the CEO, was an early employee at Ginkgo Bioworks and, during her PhD, switched from engineering machines to engineering bacteria. She has published first-author papers in Cell and Nature Biotechnology. Wang spent six years at Harvard Medical School before also joining Ginkgo. Their shorthand for the company's focus is blunt: AI-guided enzyme engineering is in their wheelhouse.

There is a small tell in the name, too. Pando is one of the largest and oldest living organisms on Earth - a single clonal aspen grove in Utah whose thousands of trunks share one root system. A company obsessed with iterating on biology at scale could not have picked a more on-the-nose namesake.

Where Pando fits

The enzyme and protein-engineering space is crowded with capable names - established players and a wave of AI-first startups all promising better proteins. Pando's position is narrow on purpose. It is not trying to be a drug company or a platform for everything. It is trying to be the group that can iterate on an enzyme faster and cheaper than the customer could in-house, and prove it with data from its own screen. In a hype cycle full of "AI will cure everything," that narrowness reads less like a limitation and more like a plan.

For a customer, the pitch is concrete. Bring an enzyme that is holding back a process - too costly, too slow, too finicky - and Pando's loop tries to hand back a version that behaves. For the industry, the more interesting question is whether the screening-first approach compounds the way the company believes it will. Every faster loop is a wager that it does.