The curse of dissociation in multistep chemistry. Illustration of the exponentially increasing complexity and resource requirements for a multi-step batch synthesis consisting of four possible step choices, up to 32 sequential steps. Reaction selection estimates are based on four possible reactants, five possible volumes, and five possible reaction times. Volume estimates are based on 10 ml of starting material and an additional 5 ml of reagent per step. Time estimates include preparation and sampling time and are based on 30 minutes to start each experiment and 5 minutes per reagent addition. Credit: Nature communicationsDOI: 10.1038/s41467-023-37139-y
A team of chemical engineering researchers has developed a self-driving laboratory capable of identifying and optimizing novel complex multistep reaction pathways for the synthesis of advanced functional materials and molecules. In a proof-of-concept demonstration, the system found a more efficient way to produce high-quality semiconductor nanocrystals used in optical and photonic devices.
“Progress in material and molecular discovery is slow because conventional techniques for discovering new chemicals rely on changing one parameter at a time using silhouette processes in chemistry and materials science laboratories,” says Milad Abolhasani, corresponding author of a project worker and professor of chemical and biomolecular engineering at North Carolina State University.
“If a complex chemistry involves dozens of parameters, it can take decades to develop a new target material or a more efficient way to produce a desired chemical.”
“Our system, called AlphaFlow, uses an artificial intelligence technique called reinforcement learning, which – when combined with automated microfluidic devices – speeds up the material discovery process. We’ve shown that AlphaFlow can run more experiments than 100 human chemists same period of the year, while using less than 0.01% of the relevant chemicals”.
“It effectively miniaturizes experiments and performs the same lab tasks that would require an entire liquid chemistry lab on an end-to-end, suitcase-sized experimental platform. It’s extremely efficient.”
AlphaFlow’s AI model makes decisions about which experiment to run next based on two things: the data it has developed from experiments it has already run, and what the results of the next several experiments will be.
“We use this moving window of past action steps and predicted outcomes of future action sequences to inform AlphaFlow’s decision-making. From this, AlphaFlow can calculate actions with delayed effects and also steer decision-making based on the most recent experimental results in real time,” says Amanda Volk, first author of the paper and a Ph.D. student at NC State. “Essentially, the system is able to instantly learn from and adapt to unexpected results.”
This is true whether the system is focused on discovering a new chemical or optimizing the production process for a known chemical. The difference is that, for discovery, the system tries to determine which precursors to add, as well as the best order in which to add them, in order to find a chemistry with the best performance.
While for optimization, the AI model already knows which precursors should be added and in which order. As a result, AlphaFlow’s focus for optimization is on determining the amount of each precursor required, as well as the time required for each reaction, to achieve the optimum yield in the most efficient manner.
“This integration of artificial intelligence and chemistry reduces the time required to develop new chemical elements by at least an order of magnitude,” says Abolhasani. “Think in terms of hours, rather than months or years.”
“AlphaFlow also offers new insights into fundamental chemistry,” says Volk. “For example, in a proof-of-concept demonstration, AlphaFlow developed a new means to produce a semiconductor nanocrystal with a cadmium selenide core and a cadmium sulfide shell. These nanocrystals are used in photonics and optical technologies. The new chemistry discovered by AlphaFlow fewer steps than previous chemistry discovered by man, making the process more efficient.”
“Furthermore, one of the steps that AlphaFlow eliminated was previously thought to be a key step in this type of multistep chemistry, which was surprising. The fact that we can produce the same high-quality nanocrystal without this step expands our understanding of chemistry involved’.
“Essentially, AlphaFlow showed that a step that researchers thought was critical turned out to be unnecessary,” says Abolhasani. “And he developed this more efficient chemistry, which changed what we thought we knew about the multistep chemistry of core/shell semiconductor nanocrystals, in just 30 days of continuous operation versus 15 years of academic literature.”
Currently, AlphaFlow is configured to perform experiments related to colloidal atomic layer deposition. This type of multistep chemistry is particularly challenging from an experimental standpoint because it involves many different parameters—there can be more than 40 variables to consider.
“However, AlphaFlow could be modified to conduct any series of experiments that involve performing chemical reactions in solution,” says Abolhasani.
“AlphaFlow is the first example we know of that integrates reinforcement learning with a self-driving lab,” says Volk. “And it highlights the extent to which artificial intelligence and the natural sciences can benefit from each other.”
The researchers are now seeking partners in both the research community and the private sector to begin using AlphaFlow to address chemistry challenges.
“Ideally, we’d like to get to a point where multiple AlphaFlow platforms are used to address different large-scale challenges related to energy transition and sustainability, but share data that will allow everyone to discover and develop new materials and molecules more quickly. says Abolhasani.
“AlphaFlow is open source. We believe it is important to share high-quality, reproducible, standardized, experimental data – both from failures and successes. We believe this is important, because we want to accelerate the discovery of new materials and chemical processes .”
The paper appears at Nature communications.
“AlphaFlow: Autonomous discovery and optimization of multistep chemistry using a self-driving fluidic laboratory guided by reinforcement learning” Nature communications, DOI: 10.1038/s41467-023-37139-y. www.nature.com/articles/s41467-023-37139-y
Provided by North Carolina State University
Reference: Self-driven laboratory speeds chemical discovery (2023, March 15) retrieved March 15, 2023 from https://phys.org/news/2023-03-self-driven-laboratory-chemical-discovery.html
This document is subject to copyright. Except for any fair dealing for purposes of private study or research, no part may be reproduced without written permission. Content is provided for informational purposes only.