Biological sparsity
Methods that depend on abundant signals can struggle in this setting.
SCIENCE
Fragment-native AI for sparse molecular signal.
The bottleneck in blood-based cancer diagnostics is learning biology from sparse, noisy molecular signals. MicroTrace is built around that view.
THE BOTTLENECK
Finding a needle in an ocean, then dropping most of the ocean.
Tumour-derived fragments make up a vanishingly small share of the cell-free DNA circulating in plasma. What is avoidable is how much of the remaining signal is discarded on the way to a model.
Signal lost in preparation
Conventional sample preparation can degrade or overwrite information carried on each molecule before it is measured.
Signal compressed away
What survives is averaged into hand-built summary statistics, and the model only sees the average.
OUR APPROACH
Preserve. Learn. Compress last.
Preserve the signal
We design the workflow to preserve multiple layers of information carried on each molecule through to the model.
Learn at the resolution of biology
Foundation-model methods learn representations directly from sparse molecular data rather than relying on features chosen in advance.
Compress only after learning
Compression comes after the model has learned what matters, so discarded information is chosen by evidence.
HOW WE WORK
Recursive scientific intelligence
Recursive Science Intelligence turns every experiment into a better next decision.
COLLABORATE
We welcome collaborations with clinicians, researchers, and industry partners.
We are building the study, assay, and clinical workflow with partners who work directly with molecular data, and cancer diagnostics.
Start a conversation →