Today, we are delighted to share new scientific discoveries in three different scientific disciplines, with preprints co-authored by experts from leading research institutions.
These publications mark a pivotal moment for Leap – and in the evolution of AI for scientific progress. We’ve demonstrated that our Discovery Engine works: we’re making real discoveries that scientists are excited to publish. And we’re doing it hundreds of times faster than existing methods, finding insight that would otherwise be overlooked.
Read our whitepaper to learn more about the technology behind our system.
Why does this matter?
We expect to change how science is done across the world. Typically, these kinds of studies would take months to go from data to publishable findings – but we’ve reduced that time to hours. Discovery Engine replaces the slow, unpredictable process of manual data exploration with a systematic, automated pipeline. Data goes in, and reliable, empirically validated insights come out.
This, in turn, makes R&D investment a much safer bet. Gathering data, through experiment or observation, can be a huge investment. And right now it’s a risky one – return on it is uncertain and slow to realise.
But how much more science would get done if we could systematically squeeze every drop of value out of that investment? How much more productive would the average researcher be, with Discovery Engine at their fingertips?
How many billions of dollars are currently left on the table, in novel insights or even major breakthroughs hidden in data, that we are about to unlock?
That’s why we’re so excited by these studies – they are the first published examples of the future of the scientific method. And, if you’ve got data, you can experience it for yourself: Leap is now open to industry pilots.
Contact us here – or scroll on to read about what we found and why scientists are excited.
Understanding Root System Tradeoffs in Arabidopsis
Biologists at the Institute of Plant Sciences of Montpellier (IPSiM), wanted to understand how Arabidopsis thaliana root systems balance competing pressures: how efficiently they transport resources, versus how costly they are to build.
Discovery Engine identified key factors that shift root systems along this tradeoff. It found that developmental stage, the hy5/chl1-5 genotype, and manganese availability were major influences — with manganese showing a unique effect not seen with other nutrients. These insights are incredibly important for understanding how root system architectures develop under different environmental conditions, crucial for optimising crop robustness and yield.
- “Saved me months of scrolling in Excel”, said Matthieu Platre, a biologist working at IPSiM. “You see what I cannot see.”
Distinguishing Tumour-Reactive T-Cell Receptors from Sequence Data
T-cell receptors (TCRs) drive immune recognition, but predicting which ones will bind to tumour antigens remains a major bottleneck in personalised immunotherapy.
In collaboration with researchers at the University of Washington, we used Discovery Engine to analyse published TCR sequences from cancer patients, surfacing interpretable, combinatorial rules that were predictive of reactivity.
These rules combined features like CDR3 length, net charge, and hydrophobicity, across both the alpha and beta chains. More than half of the predictive patterns involved features from both chains — an important insight that supports the growing focus on dual-chain analysis.
- "I think this is the next frontier of bioinformatics”, said Emma Bishop, a bioinformatician at the University of Washington. “I wish every paper was this easy.”
Read the paper: Automated Discovery of Patterns in T-Cell Receptor Physicochemical Signatures
Disproving a Foundational Assumption in Meteorology
Monin–Obukhov Similarity Theory (MOST) underpins nearly all surface flux estimates in the atmospheric boundary layer. But in coastal marine settings, it turns out that its assumptions often don’t hold.
In partnership with the National Center for Atmospheric Research (NCAR), we applied Discovery Engine to flux data collected from offshore buoys as part of the Coastal Land–Air–Sea Interaction (CLASI) project.
Discovery Engine found that wind speed decreases with height in nearly 20% of observations, and surfaced large vertical gradients in sensible heat flux near the sea surface — both in contradiction to MOST. This finding has huge implications for meteorological modelling, improvements in which are valued at billions of dollars.
- “It would take us one postdoc year to analyze this data”, said Patrick Hawbecker, Project Scientist at NCAR. “And [Discovery Engine] found something that we may never have found, that could be worth billions.”
Read the paper: Explaining Surface Layer Theory Departures in Marine Flux Profiles with Data-Driven Discovery
What’s next?
For centuries, the pace of discovery has been limited by human attention: researchers manually exploring their results, guided by intuition and constrained by time. This was fine four hundred years ago, when experimental and observational data was still on a human scale – but now, it’s wildly underpowered and the biggest bottleneck to scientific progress. With Discovery Engine, this bottleneck disappears. Scientists can now test thousands of hypotheses in parallel, surface the most promising (and surprising) leads, concretely identify the best next experiment and do all of this 100x faster than before.
This is just the beginning. From agriculture to immunotherapy, from environmental modelling to materials science, every discipline stands to benefit. The scientists we’ve partnered with are already publishing faster, learning more, and rethinking what’s possible in their disciplines.
