August 25, 2022

Acceleration Consortium, Matter Lab, and Vector Institute collaborate on software to power self-driving labs

We've teamed up with the Matter Lab and the Vector Institute to develop Gryffin, an open-source “off-the-shelf” software to help power self-driving laboratories (SDLs) which combine AI, robotics, and advanced computing to reduce the time and cost of bringing materials to market.


The Acceleration Consortium (AC), the Matter Lab, and the Vector Institute have teamed up to develop Gryffin—an “off-the-shelf” software package to power the back-end of autonomous laboratories that will ultimately propel the next generation of scientific experimentation.

The code is open-source and available for download.

An artificial intelligence (AI) revolution is well underway in materials science. AI and machine learning have the potential to disrupt current processes, as they become integrated into the automated design-make-test experimentation cycle of functional molecules and advanced materials via an emerging technology called self-driving labs (SDLs).

While traditional materials innovation processes are often prohibitively slow and expensive, SDLs combine AI, robotics, and advanced computing to radically reduce the time and cost of bringing materials to market, from 20 years and $100 million to as little as one year and $1 million. Gryffin has already been employed to drive several prototypes of SDLs, targeting diverse applications including light-harvesting materials, chemical reaction optimization, and nanotechnology.

“The AI community in Toronto and Canada is burgeoning, as opportunities to apply AI continue to multiply across industries,” says Deval Pandya, the Vector Institute’s Director of AI engineering. “This collaboration between Vector, Matter Lab, and the AC will be one of many, as the application of AI for discovery continues to expand and gain new adopters. The future of discovery will be digital and data-driven.”

“By taking an open innovation approach, we believe there is greater potential to scale this technology, and to do so quickly, says Alán Aspuru-Guzik, Director of the Acceleration Consortium. To usher in an era of “materials on demand,” the technology must be ubiquitous and easy to use. Gryffin gets us a step closer to one day developing a modular, affordable, out-of-the-box self-driving lab, containing not just the software but the synthesis and characterization robots as well.”

How it works

Using AI and computational modelling, Gryffin predicts which combination of materials or molecules will have the desired properties (e.g.,conductivity) before a robotic lab synthesizes and tests it for said properties. This data is then fed back into the AI system, so that it can learn from the results to generate a new, better slate of candidates. After rounds of predictions, syntheses, and tests, a winner emerges.

The process is similar to cooking, an analogy recently made by Jason Hattrick-Simpers, the AC’s Associate Director of Data and Research Networks, in University of Toronto Magazine. A meal can turn out wildly different depending on the order in which you add each ingredient or whether you fry, bake, or boil them. In materials science, there is the same sort of problem. There are infinite ways in which these elements could be put together. “Data-driven optimization software like Gryffin helps researchers efficiently determine the optimal recipe to achieve the material they desire,” explains Riley Hickman, one of the Matter Lab’s lead researchers on the project. Hickman worked closely with the Vector Institute's John Willies, who was the project's tech lead.

More about Gryffin

Gryffin is a general-purpose optimization framework for the autonomous selection of categorical (e.g., solvent, catalyst, ligand, etc.) and mixed continuous-categorical parameters (e.g., temperature, reaction time, concentration, etc.) driven by expert knowledge. This linear-scaling Bayesian optimizer uses a kernel regression surrogate, natively supports batched optimization, and allows for specification of intuitive biases between explorative and exploitative sampling behaviour. The algorithm is also capable of leveraging expert knowledge in the form of physicochemical descriptors of categorical options to further accelerate its optimization rate.

About the partners

Founded in 2017, the Vector Institute works with industry, institutions, startups, and governments to drive research excellence and leadership in AI to foster economic growth and improve the lives of Canadians. Vector is funded by the Province of Ontario, the Government of Canada, and industry sponsors.

Based at the University of Toronto, the Acceleration Consortium (AC) is a global community of academia, government, and industry who are accelerating the discovery of new materials and molecules needed for a sustainable future.

The Matter Lab at the University of Toronto accelerates the discovery of new chemicals and materials that benefit society using new technologies such as quantum computing, machine learning, and automation.

Both the AC and the Matter Lab are led by Alán Aspuru-Guzik, who is also a CIFAR AI Chair and Faculty Member at the Vector Institute.


Acceleration Consortium

This piece was written by a member of the AC team

Deadline to apply for our 2023 AC postdoc fellowship is Jan 27!