Keep pulling the thread on Dr. Andrew Hopkins.
Xyme has built proprietary quantum chemistry datasets that are 10,000 times larger than what is available in the public domain.
Xyme has rewritten underlying code for quantum chemistry calculations, reducing computation times for methods like density functional theory from months on supercomputers to seconds.
Xyme's generative AI designed a novel lipase enzyme on its first attempt that is so different from known sequences it would have taken nature an estimated one billion years to evolve.
In Xyme's proof-of-concept for enzyme design, 90% of the AI-generated candidates tested were active, a hit rate described as "unheard of" in the field.
Xyme models a pathway to producing a second-generation liquid sustainable ethanol fuel at a cost of only 25 cents per litre.
Xyme is developing a single-reaction process to treat bio-oils, which would allow oil refineries to increase the blend of sustainable oils with fossil fuels from the current 1% limit to as high as 30%.
Lord John Browne, the former head of BP, is a co-founder and chairman of Xyme.
Xyme expects to have its first products ready for market by the end of 2027.
Xyme's approach to manufacturing using biological molecules parallels the methodology used by Google DeepMind for its AlphaFold protein folding model.
Xyme is developing a new type of AI that understands the physics of processes to achieve true invention, moving beyond the descriptive capabilities of current large language models.
The data required to train Xyme's physics-based AI models is not available on the public internet and must be generated from sources like quantum chemistry calculations.
Xyme utilizes flow matching generative AI architectures because they allow for the embedding of models within other models, such as placing quantum AI models inside a generative framework.