Locus

Locus turns failed biological experiments into actionable intelligence. Today, scientists run complex experiments that can take months and cost thousands of dollars, yet when they fail, it’s hard to understand why. With so many interacting variables shaping biological outcomes, no system or person can reliably pinpoint the causes of failure. We are building a biological debugging engine that transforms experimental results into actionable insight, starting with CRISPR, where failures are frequent and costly. Locus aims to create a new mode of experimental insight—compressing research timelines, reducing wasted effort, and ultimately, accelerating the pace at which biological innovation improves human life.

Failure is the backbone of science. It’s an essential part necessary for progress and discovery. But there’s a difference between productive and repetitive failure. We know this because we’ve experienced it. Last semester, when transfecting bioengineered DNA into cells to reduce cancer signaling, I, Mahilan, ran through a month-long pipeline—digestion, PCR, Gibson assembly, transformation, etc., and then imaged the cells and ran the analysis. Did cancer signaling decrease? No. More strikingly, the transfection itself failed. Why? No clear answer. My mentor’s advice: try it again. This might sound crude, even unscientific. But it’s the reality of biological research. With dozens of interacting variables—many we can’t fully observe or control—pinpointing the cause of failure is incredibly difficult. Scientists make informed guesses and iterate. Trial-and-error isn’t a flaw in the system; it is the system. The problem is that biology has no way to systematically learn from these failures. Unlike software, where a failed execution produces a traceable error log, a failed experiment leaves only an outcome, with no clear causal explanation. Existing tools like Benchling and SnapGene are powerful, but they stop at documentation. They record what happened, but don’t reason about why. As CRISPR adoption accelerates and workflows grow more complex, this gap becomes more expensive. More weeks lost and drug discovery timelines stretched. In a market projected to reach $24 billion in the next decade, this is a structural bottleneck. Locus intervenes by doing something no tool in biology has ever done: computationally modeling experimental workflows themselves. Picture this: you're reading a protocol, wondering what would change if you altered a reagent or adjusted a step. Today, that means running the experiment. With Locus, it doesn't. We model workflows as structured systems—each step a node with tunable parameters, drawing on techniques from formal program analysis. Before you run, abstract interpretation flags high-risk steps and likely failure points. After failure, symbolic execution reconstructs the path, compares it to successful runs, and isolates the minimal causal deviation. Trained on thousands of experimental workflows from the literature, Locus identifies patterns no individual researcher could track. Where scientists guess, Locus learns—turning every experiment into a system that improves the next. Importantly, Locus is not only technically sound but also makes sense from a business lens. First, while most AI in biology treats failed experiments as noise in the data, we treat them as signals. Put simply, every failure and new lab that joins strengthens the model. That flywheel is a definitive moat. Second, Locus deepens over time. When labs run experiments through Locus, they build proprietary experimental intelligence tied to their own workflows. Switching means finding a new tool and losing valuable insight you can’t recreate. Third, the market is immediate. We start with roughly 1,000 CRISPR-focused biotechs in the U.S., each losing around $50K annually to undiagnosed failures. But CRISPR is the wedge. Every biological experiment is a protocol waiting to be debugged. Locus is building the reasoning layer biology has never had: where each experiment makes the next one smarter.