R&D teams are losing trillions in failed experiments due to their inability to fully explore their data. We enable this exploration increasing the speed, frequency, and novelty of scientific breakthroughs.

Traction:



• 2 novel discoveries
• 11 ongoing collaborations in a variety of sciences
• First industry pilot negotiations

Society is investing more than ever in research – but scientific progress is slowing. Most papers are wrong. Ideas are getting harder to find.

Our ability to generate scientific data is not matched by our ability to make sense of it.

Scientists leave trillions of dollars on the table – in potentially novel insights or even major breakthroughs hidden in their data – simply because no one thought to ask the right questions.

Scientists are limited to testing only the hypotheses they can think of – driven by intuition, assumptions, and papers that often don’t replicate.

Even the best analytics tools cannot capture complex, non-linear relationships – unless they already know what to look for.

This means scientists access only a tiny fraction of the possible discoveries in their data – and understanding that fraction can take months of manual analysis.

Language models inherit and amplify these problems. They’re trained on the same flawed material, to automate the same limited process.

We train AI models on scientific data – and then use interpretability to understand what they’ve learned.

This results in novel discoveries at unprecedented speed and scale.

• 100x faster than existing methods
• Systematic and unbiased
• Data efficient
• Works in any domain!

This general purpose technology is fundamentally changing how science is done.

Our key insight: deep neural networks find meaningful patterns in data that humans miss. And with our proprietary interpretability algorithms we are able to automatically extract those patterns, contextualise them and render them human parseable.

This gives us the unique ability to use deep learning’s superhuman pattern recognition to find novel insights in data, in a completely unbiased way.

We have already made novel discoveries. In plant biology, we discovered a novel combination of genotype and nutrient profile that dramatically improves root structures – crucial for improving crop robustness and yield (https://www.leap-labs.com/blog/discovering-hidden-patterns-in-plant-biology).

In meteorology, we found strong evidence disproving a foundational assumption of the field – this has massive implications for meteorological modelling, improvements in which are valued at billions of dollars (https://www.leap-labs.com/blog/surfacing-the-unexpected-improvements-in-meteorological-modelling.

Our method is domain-agnostic, but we are beginning with two verticals: AgBio and Materials Science. AgBio is a large market and provides an onramp to the more difficult to enter BioPharma market. The data is similar, and many large bio companies like Bayer have both crop and pharma science divisions. Materials science is applicable to many industries including energy, chemicals, biotech, aerospace, and manufacturing. Our first mat sci focus is in energy.

leap-labs.com