Enveda’s drugs from plants will turn pharma upside down

It’s easy to be skeptical when you first hear about Enveda and its 30-something founder Viswa Colluru. The business he’s in – making medicines from plants and other natural products – is filled with charlatans. And Colluru lately has been making not just bold claims but ones that seem flatly ridiculous.

He says his five-year old biotech company has found a way to develop medicines from nature four times faster and for a tenth the cost of traditional pharmaceuticals.

Sure, the earliest medicines came from natural compounds. A small handful still do. I have personal experience with one of them. But most drugs are researched and developed in labs. And they are developed that way for a reason.

Drug development relies on compounds that can be predictably and reliably researched and manufactured at scale. Plant biology and supply are full of variables. Plants are also trickier to patent. This typically makes the already expensive and time consuming drug development process even more so.

Indeed, “going natural” has become something of a third rail in drug development. Dozens of pharma companies big and small have tried. Few have succeeded. It’s true that nature is filled with undiscovered drug compounds that could become revolutionary medicines worth billions. But most in the pharma industry have come to believe it would cost tens of billions to find and develop them.

But dig into what Colluru and Enveda are building, and you find an effort that’s not only legitimate but one that could very well deliver on its outlandish claims.

It already has a plant compound in a phase II clinical trial for atopic dermatitis and the same compound in a phase II trial for asthma. It has two others – to treat obesity and to treat irritable bowel disease – in phase I trials. And it has another dozen drug candidates in earlier stages of development. The dermatitis/asthma drug is derived from a plant that grows in Southeast Asia. Enveda won’t say which one. The obesity compound is derived from human blood. 

If future steps go as well as its previous trials, Enveda could have four FDA approved medicines in three to five years – eight to ten years after its founding. That’s incredibly fast. It typically takes seven to ten years and more than $1 billion to take just one compound from candidate to FDA approval.

All this has investors salivating. Enveda raised $150 million in September 2025. Mikael Dolsten, Pfizer’s former long-time chief scientific officer, joined the board. It’s raised more than half a billion since Colluru co-founded the company in 2019. Sanofi, the French pharma giant, invested $20 million in early 2025. The September fund raise valued the company at more than $1 billion, giving it unicorn status.

Yes, the approval rate even for drugs that make it into phase III trials is only about 50 percent. But just the speed that Enveda is finding and testing drug candidates has investors betting Enveda’s new approach will lead somewhere interesting even if these first drugs fail.  “I know of no drug company that after five years has four drugs in clinical trials with only 200 people,” said Rohit Sharma, a partner at True Ventures and one of Enveda’s earliest backers.

All of it has made Enveda not just an interesting new company to watch but something of a test case for the transformative powers of both AI and robotics in biotech. Only about one in ten drug development candidates get approved. Many believe that AI used in some new way will soon help increase that approval rate and, by extension, meaningfully lower development costs. But there hasn’t yet been much hard evidence demonstrating exactly how that will happen.

Enveda says its secret sauce is the way it leverages mass spectrometry data, AI and robotics.  At its 60,000-square-foot labs at its Boulder, CO headquarters and in Hyderabad, India, mass spectrometers the size of giant refrigerators analyze plant extracts to create unique molecular fingerprints. Enveda’s proprietary AI processes that data to identify what those molecules do. Then, dozens of lab robots – some small enough to fit inside the case of a desktop printer – conduct thousands of trial and error tests to narrow down which molecules work on the disease targets. 

Colluru said that it’s the trial and error rate that this technology combination is able to achieve that is critical to Enveda’s success. “It would take forever with (a lab staffed) with humans,” he said.  “What we’ve done is trial and error at scale and at cents on the dollar compared with others.”

Colluru said he saw the connections among these not obviously connected technologies after reading a 2018 paper in the Journal of Natural Products by Pieter Dorrestein, a professor of pharmacology and pharmaceutical sciences at UC San Diego. He runs its Collaborative Mass Spectrometry Innovation Center.  

Dorrestein had been trying to leverage the molecular fingerprint of mass spectrometry data for about half a decade without success. He thought it could be the foundation of something that resembled Google search, but for molecules.

Colluru had a much bigger idea than just a search engine.  “The first time I met Viswa was when he came to my office with one of my papers in 2019 saying ‘I want you to help me build a company from this,’” Dorrestein said. He is now Enveda’s scientific co-founder.

It’s still too soon to know whether Enveda’s approach will live up to Colluru’s boasts – drug development at a tenth the cost and four times the speed. But even if they get close it could be one of the more important developments of the 21st Century. And  the time Enveda has spent collecting and training its AI systems – two years longer than the three-year-old AI revolution itself – suggests they have a meaningful head start on competitors. 

First mover advantage in a new marketplace often proves much less valuable than originally believed. Google was one of the last search engines, for example. But in the AI revolution first mover advantage is proving to be critical. 

That’s because AI systems are only as good as the training data they consume and the time its trainers have to tweak their outputs and reprocess them. The more time you have to iterate the better.

Why? Because once that iteration process hits a certain point, it seems to generate a flywheel effect – exponentially large improvements with each iteration. It’s why we are seeing such accelerating increases in capabilities of AI chatbots right now. 

This flywheel, Colluru says, “is our longest enduring moat.”

Here is our conversation, edited for length and clarity:

Fred Vogelstein: Where did the idea to start Enveda come from?

Viswa Colluru: It goes back to me losing my mother as a 13-year-old to chronic myeloid leukemia. She was diagnosed in the late 90s. I lost her December 24, 2003. So that was the big compass.

It led me to study biology in undergrad. It was a subject I barely passed in high school. And that gave me the courage to do biomedical research after that. This was something I never realized I had either the attitude or the aptitude for.

So I moved from sunny south India to (the University of Wisconsin) in Madison where it’s so cold in the winter – minus 40 degrees sometimes – that the Celsius and Fahrenheit scales converge.

I learned a lot about the scientific enterprise and asking arcane but profound questions about the immune system. But my biggest lesson was this: We frequently confuse what’s new with what’s exciting. I would humbly posit that more often than not, inventions and technologies that change the fabric of our lived experience are not new. They’re an old idea that was unfinished.

Immunotherapy was unfinished.  Rapamycin was unfinished. Transplant medicine was unfinished. Statins were unfinished. Every one of those things took sometimes the lifetime of one inventor and often longer before the ideas went from seeming crazy to being indispensable.

So while the world is always harboring a heady obsession for what’s new, I decided I was going to look for what is old but unfinished. AI and computer science actually felt old in 2016. And nobody wanted to believe that marrying computer science and the life sciences was going to be useful.

FV: You worked at Recursion in Salt Lake City for three years until 2019. What did you see there that made you confident you could start Enveda?

VC: Recursion was extremely bold. And that was definitely reinforcement for me. But what Recursion also did – that everybody else did – was to just go after biology. “Here is this biological thing. Let’s find the chemistry to drug it with. I’m going to anoint a protein that I think is the culprit towards the pathogenesis of this disease. Now all I need is a molecule that stops the protein from doing what it does.”

What this does is that it relegates chemistry to the level of a tool. It makes biology all that matters. And once you relegate chemistry to a tool, you’re unwilling to invest anything meaningful in the best chemistry.

FV: Can you be more specific about what you mean?

VC: A key part of why we’re succeeding today is because we have a completely different approach.

It turns out that the way that molecules break apart in a mass spectrometer – the device metabolomic scientists use to get these fingerprints – is actually really similar to how humans weave together words in a sentence.

The presence or absence of a fragment (word) doesn’t on its own mean anything unless there are other fragments (words) surrounding it. That’s essentially the way the AI transformers technology works.

FV: So you use transformer AI (the same AI behind the current AI revolution) to spot patterns in all this data that would take humans a lifetime to spot?

VC: Correct. So that was a key breakthrough. Then through a whole series of other innovations involving robots moving tiny amounts of liquid from plate to plate and using fancy statistics we can pinpoint which molecule is the one that’s driving the activity.

That’s what we do at scale. That allows us to answer the second question: What are those molecules doing in those samples?

FV: How is this different from the way things were done previously?

VC: The approach that companies have taken for the last generation was to sequence the genes present in diverse natural organisms and try and divine the molecules that they made. So you essentially ended up on a long scientific project to get one single compound. It was a genome centric exploration of the chemistry of the world.

The problem is that cloning genes out of plants is really difficult. Plants have extremely clumped, complex genomes. They can have dozens of copies of a single gene.

The chemistry of plants is so complicated because they can’t solve their problems with movement. You and I can walk away from a problem. Plants cannot. So they do it through changing their chemistry. That affects the microbes that grow and don’t grow on them, the bees that pollinate or don’t pollinate, and the insects and herbivores that feed on them.

And so essentially this entire universe of plant chemical space had been left out of the modern scientific revolution and modern approaches to drug discovery.

FV: What led you to the idea of mixing AI with mass spectrometry data?

VC: I read a lot of papers and learned about a field called metabolomics. That’s the study of an organism’s metabolism. It turned out that one of the big topics in metabolomics was whether we could come up with a unique fingerprint for all of the molecules in a sample.

And I learned about the work of our scientific co-founder Pieter Dorrestein, who had created the first database of these fingerprints (at UCSD’s Skaggs School of Pharmacy and Pharmaceutical Sciences).

The problem was that those molecular fingerprints historically were really difficult for people to interpret. So they were only used essentially as a fingerprinting database.

That creates the same problem real fingerprint databases have today: If there is no fingerprint match in the database, that’s a dead end. That was the state of the art of metabolomics when Pieter and I and Enveda started working on it.

So now we can take 100 plants that people from all over the world have used to treat diseases and symptoms, bring those into our lab, extract all the chemistry and identify the unique molecules.

Then we can turn them into medicines that are not only manufactured better but are a more effective treatment than the original plant. We do this at a fraction of the time and the cost of your average biotech company.

So today if I take a vial of your blood, we can tell you about 10 times more about the molecules in it than any other investigative team in the world. And we want to be able to tell you 100 to 1,000 times more in the next five years in terms of the kinds of molecules that are present in the sample.

FV: Where do all these plants come from? Do you have a team of people who come back with bags of random plants from the jungles of the world?

VC: We actually want to send expeditions to the Amazon and all of these other places very soon. But to give us a healthy start, there are about 38,000 plants that have publicly recorded uses across dozens of cultures that we’ve absorbed into a database.

We use that to pick what plants may have desirable molecules that science has not yet uncovered.

FV: How do you patent drugs made from molecules that occur naturally and are in a public database?

VC: We modify them so that they’re no longer purely natural, but derived and inspired by nature. Salicylic acid is what’s found in nature in the bark of a willow tree. Acetylsalicylic acid is aspirin. Doing that modification is what allows us to make them more manufacturable, more dosable, more stable, more effective, and potentially more safe.

FV: Help me visualize the process. You have labs in India and Boulder. What does the process look like inside?

VC: The visual of it is the mass spectrometers take an extract of a plant, break each individual molecule down to create that fingerprint. Our AI – the world’s best software for this kind of data – reads that. Then we use that information to identify what each of these molecules do.

We split the sample into thousands of wells – little experimental vesicles on a plate – with cells that are, for example, inflamed.

It used to take a long time to do enough tests to identify which subset of molecules make the cells less inflamed. So we built a bunch of custom hardware that’s very good at separating these extracts and identifying which exact molecules were put in each well. And so that’s the robotics part of it.

FV: So it’s a giant trial and error experiment that would take forever with rows of lab techs but is very cost effective if done with robots?

VC: It would take forever with humans. And that was why finding medicines from the natural world was always so unpredictable and risky. The rate of testing these molecules and discovering their functions was too slow.

What we’ve done is trial and error at scale and at cents on the dollar compared with others. Over the next year or so, we’ll have 19 different clinical signals from our first four medicines that we’ll be getting back. So we’re on the cusp of seeing if we’re taking a very old idea and reimagining it with technology.

FV: Is the phase one trial process shorter and faster because you are using natural compounds? 

VC: The bar for you to generate evidence for safety is the same: Same number of patients, same length of trial, et cetera. However, because we’re starting with molecules where nature has pre optimized them to have biological utility for billions of years, the process where we go from the original molecule to a nature derived molecule that we own, we can do four times faster than the industry.

So we can go from the first inkling – it’s called a biological hit – to a development candidate in a year. The industry takes about four and a half to five and a half years on average.

FV: Why is that?

VC: Because nature’s done all the work. Nature’s done three fourths of the work of making it compatible with being in a cell.

Synthetic compounds that are just synthesized by a human chemist are not necessarily optimized to be compatible with mitochondria, to be compatible with living membranes, to be compatible with DNA.

We don’t know exactly why our molecules from nature come out pre-baked – being ready to actually have biological utility without having biological toxicity – but we are thoroughly unsurprised.

Usually molecules fail these toxicology tests about 50% of the time. We have now tested five molecules and, touch wood, every one of them has passed safety testing with flying colors.

Is there something to suggest these molecules are safer? Probably. But is it accurate to suggest that all natural molecules are safe? No. Some of the world’s most dangerous molecules are natural.

FV: So … what’s your competitive advantage? Your source materials are plants that have known therapeutic effects. What would stop a competitor from saying, “I’ll do that too”?

VC: Because we’ve been doing this five years already, and, as a result, have the world’s best and largest proprietary data sets that power our algorithms. We created those datasets by running lots and lots of experiments. We’ve run mass spectrometers 24 hours a day 7 days a week to create three times the entire public database of mass spectrometry.

And that’s our longest enduring moat. Because our algorithms are better, we know which data to collect to make them better than everyone else’s.

FV: Thank you for spending so much time explaining this.

DIsclosure: Enveda is backed by multiple investors including True Ventures where my partner Om is a partner emeritus. While he is part of the firm, he is not involved with the company directly or indirectly.