Integrated Biosciences

Here is a thing about drug discovery, which is that it is mostly a data problem wearing a biology costume. You want to find a small molecule that does something useful inside a cell. To do that, you need clean examples of the cell doing the thing, and clean examples of it not doing the thing, and enough of both that a computer can tell them apart. The trouble is that cells are noisy, uncooperative, and generally uninterested in producing tidy datasets for your machine-learning model.

Integrated Biosciences, founded in 2022 and based in Redwood City, has a proposed fix that is either very clever or very obvious depending on how you feel about optogenetics. Optogenetics is a technique from neuroscience where you engineer cells to respond to light, so you can flip a biological pathway on and off with a flash - at millisecond speed, at micron resolution. It was built to control neurons. Integrated Biosciences pointed it at drug screening instead.

The pitch, as co-founder Max Wilson has put it, is not subtle: "Drug discovery is broken, and we are here to fix it." What I appreciate about that line is that the company actually defined what "broken" means - noisy data, undruggable targets - and then built a specific tool for each complaint. That is more discipline than the average "AI for X" startup, which tends to define the problem as "there is not enough AI in X."

So the loop goes like this. Use light to precisely control a cellular pathway. Because you controlled it precisely, you get unusually clean data about what the cell does. Feed that data to a proprietary AI engine. The AI proposes small molecules. Then - and this is the part they insist on - the AI explains why the molecule should work, pointing at the structural reason rather than shrugging and saying the model liked it. Then chemists make the molecule. It is a real pipeline, and each step feeds the next.

The target market is aging, which is a strange thing to call a market. Aging is not one disease; it is the largest single risk factor for a long list of them. The bet is that if you can drug the underlying biology of aging, you address many conditions at once. This is either the most important idea in medicine or a very expensive way to run a lot of assays, and the honest answer is that we do not yet know which. But the company has already done something concrete with the platform, which brings us to the antibiotics.

In December 2023 the company announced a paper in Nature describing a new structural class of small-molecule antibiotics. This matters for two reasons that are worth separating. The first is that new antibiotic classes are genuinely rare - this was among the first in roughly 60 years, at a moment when antibiotic resistance is a slow-motion public-health emergency. The second reason is more about method than molecule.

The compounds were found using explainable deep learning. The distinction between explainable AI and the more common black-box variety is not academic. A black-box model tells you a molecule is promising and declines to say why, which means when it is wrong you have no idea where it went wrong, and when it is right you cannot reliably repeat the trick. An explainable model points at the structural features driving its prediction. That turns a lucky hit into something closer to a repeatable engine - which is, after all, what a drug-discovery platform is supposed to be.

So the antibiotics work is less a product than a demonstration. It says: this platform can find real, novel, publishable chemistry, and it can tell you why the chemistry works. The company's actual ambition is aging, but antibiotics were a convenient way to prove the machinery runs. Research from the team has appeared in Nature, Nature Aging, and Cell Systems.