Erin Warner

In less than a month, researchers have used AlphaFold, an artificial intelligence (AI)-powered protein structure database, to design and synthesize a potential drug to treat hepatocellular carcinoma (HCC), the most common type of primary liver cancer.

The researchers successfully applied AlphaFold to an end-to-end AI-powered drug discovery platform called Pharma.AI. That included a biocomputational engine, PandaOmics, and a generative chemistry engine, Chemistry42. They discovered a novel target for HCC – a previously undiscovered treatment pathway – and developed a “novel hit molecule” that could bind to that target without the aid of an experimentally determined structure. The feat was accomplished in just 30 days from target selection and after only synthesizing seven compounds.

In a second round of AI-powered compound generation, researchers discovered a more potent hit molecule – although any potential drug would still need to undergo clinical trials.

The study – published in Chemical Science – is led by the ��Ƶ Acceleration Consortium Director Alán Aspuru-Guzik, Nobel laureate Michael Levitt and Insilico Medicine founder and CEO Alex Zhavoronkov.

“While the world was fascinated with advances in generative AI in art and language, our generative AI algorithms managed to design potent inhibitors of a target with an AlphaFold-derived structure,” Zhavoronkov said.

“AlphaFold broke new scientific ground in predicting the structure of all proteins in the human body,” added co-author Feng Ren, chief scientific officer and co-CEO of Insilico Medicine. “At Insilico Medicine, we saw that as an incredible opportunity to take these structures and apply them to our end-to-end AI platform in order to generate novel therapeutics to tackle diseases with high unmet need. This paper is an important first step in that direction.”

AI is revolutionizing drug discovery and development. In 2022, the AlphaFold computer program, developed by Alphabet’s DeepMind, predicted protein structures for the whole human genome – a remarkable breakthrough in both AI applications and structural biology. This free AI-powered database is helping scientists predict the structure of millions of unknown proteins, which is key to accelerating the development of new medicines to treat disease and beyond.

Scientists have traditionally relied on conventional trial-and-error methods of chemistry that are slow, expensive and limit the scope of their exploration of new medicines. As COVID-19 has demonstrated, the speedy development of new drugs or new formulations of existing ones is needed – and increasingly expected by the public. AI has the potential to deliver this speed by transforming materials and molecular discovery, as it has done with just about every branch of science and engineering over the last decade.

“This paper is further evidence of the capacity for AI to transform the drug discovery process with enhanced speed, efficiency, and accuracy,” said Michael Levitt, a Nobel Prize winner in chemistry and the Robert W. and Vivian K. Cahill Professor of Cancer Research and professor of computer science at Stanford University. “Bringing together the predictive power of AlphaFold and the target and drug-design power of Insilico Medicine’s Pharma.AI platform, it’s possible to imagine that we’re on the cusp of a new era of AI-powered drug discovery.”

Both Insilico Medicine – a clinical stage company that counts both Aspuru-Guzik and Levitt as advisers – and U of T’s Acceleration Consortium are working actively to develop self-driving laboratories, an emerging technology that combines AI, automation and advanced computing to accelerate materials and molecular discovery. Accessible tools and data will help more scientists enter the field of AI for science, in turn helping to drive major progress in this area.

“What this paper demonstrates is that for health care, AI developments are more than the sum of their parts,” said Aspuru-Guzik, a professor of chemistry and computer science in U of T’s Faculty of Arts & Science and the Canada 150 Research Chair in Theoretical and Quantum Chemistry. “If one uses a generative model targeting an AI-derived protein, one can substantially expand the range of diseases that we can target. If one adds self-driving labs to the mix, we will be in uncharted territory. Stay tuned!”

'No small feat': U of T's Anatole von Lilienfeld is using AI to explore the vastness of 'chemical space'

Christopher.Sorensen — Mon, 05 Dec 2022 16:55:07 +0000

'No small feat': U of T's Anatole von Lilienfeld is using AI to explore the vastness of 'chemical space'

Christopher.Sorensen Mon, 12/05/2022 - 11:55

Cutline

Anatole von Lilienfeld is one of the world's brightest visionaries on the use of computers to understand the vastness of chemical space. (photo by Diana Tyszko)

Erin Warner

Topic

Our Community

Acceleration Consortium

Institutional Strategic Initiatives

Artificial Intelligence

Department of Chemistry

Faculty of Applied Science & Engineering

Faculty of Arts & Science

Rersearch & Innovation

Vector Institute

The ��Ƶ’s Anatole von Lilienfeld navigates space – but rather than exploring the depths of the universe, his artificial intelligence-powered work focuses on “chemical space” and the untapped potential of undiscovered chemical combinations.

The inaugural Clark Chair in Advanced Materials at U of T and the Vector Institute for Artificial Intelligence – and a pivotal member of U of T's Acceleration Consortium – von Lilienfeld is one of the world's foremost visionaries for the use of computers to understand the vastness of chemical space.

Von Lilienfeld, a professor jointly appointed to U of T’s department of chemistry in the Faculty of Arts & Science and the department of materials science and engineering in the Faculty of Applied Science & Engineering, was a speaker at the Acceleration Consortium’s first annual Accelerate conference earlier this year. The four-day program explored the power of self-driving labs, an emerging technology that combines AI, automation and advanced computing to accelerate materials and molecular discovery.

Writer Erin Warner recently spoke with von Lilienfeld about the digitization of chemistry and what the future holds.

How big is chemical space?

We are surrounded by materials and molecules. Consider the chemical compounds that make up our clothing, the pavement we walk on, and the batteries in our electric cars. Now think about the new possible combinations that are out there waiting to be discovered, such as catalysts for effective atmospheric CO2 capture and utilization, low-carbon cement, lightweight biodegradable composites, membranes for water filtration, and potent molecules for treatment of cancer and bacterial-resistant disease.

In a practical sense, chemical space is infinite and searching it is no small feat. A lower estimate says it contains 10⁶⁰ compounds – more than the number of atoms in our solar system.

Why do we need to accelerate the search for new materials?

Many of the most widely used materials no longer serve us. Most of the world’s plastic waste generated to date has not yet been recycled. But the materials that will power the future will hopefully be sustainable, circular, and inexpensive.

Conventional chemistry is slow, a series of often tedious trial and error that limits our ability to explore beyond a small subset of possibilities. However, AI can accelerate the process by predicting which combinations might result in a material with the set of desired characteristics we are looking for (e.g., conductive, biodegradable, etc.).

This is but onestep in self-driving laboratories, an emerging technology that combines AI, automation, and advanced computing to reduce the time and cost of discovering and developing materials by up to 90 per cent.

How can human chemists and AI work together effectively?

AI is a tool that humans can use to accelerate and improve their own research. It can be thought of as the fourth pillar of science. The pillars, which build on each other, include experimentation, theory, computer simulation and AI.

Experimentation is the foundation. We experiment with the aim of improving the physical world for humans. Then comes theory to give your experiments shape and direction. But theory has its limitations. Without computer simulation, the amount of computation needed to support scientific research would take far longer than a lifetime. But even computers have constraints.

With difficult equations come the need for high-performance computing, which can be quite costly. This is where AI comes in. AI is a less costly alternative. It can help scientists predict both an experimental and computational outcome. And the more theory we build into the AI model, the better the prediction. AI can also be used to power a robotic lab, allowing the lab the ability to run 24/7. Human chemists will not be replaced; instead, they can hand off tedious hours of trial and error to focus more on designing the objectives and other higher-level analysis.

Are there any limitations to AI, like the ones you described in the other pillars of science?

Yes, it is important to note that AI is not a silver bullet, and that there is a cost associated with it that can be measured in data acquisition. You cannot use AI without data. And data acquisition requires experimenting and recording the outcome in a way that can be processed by computers. Like a human, the AI then learns by reviewing the data and making an extrapolation or prediction.

Data acquisition is costly, both financially and in terms of its carbon footprint. To address this, the goal is to improve the AI. If you can encode our understanding of physics into the AI, it becomes more efficient and requires less data to learn but provides the same predictive qualities. If less data is needed for training, then the AI model becomes smaller.

Rather than just using AI as a tool, the chemist can also interrogate it to see how well its data captures theory, perhaps leading to the discovery of a new relative law for chemistry. While this interactive relationship is not as common, it may be on the horizon and could improve our theoretical understanding of the world

How can we make AI for discovery more accessible?

The first way is open-source research. In the emerging field of accelerated science, there are many proponents of open-source access. Not only are journals providing access to research papers, but also in many cases to the data, which is a major component for making the field more accessible.

There are also repositories for models and code, like GitHub. Data sets can record and encode a lot of value.Providing more open access to data, which can be too costly for some to generate on their own, could lead to scientific advancements that ultimately benefit all of humanity. Scientists can then use the data from other scientists to ask their own research questions and make their own AI models.

A second way to expand AI for discovery is to include more students. We need to teach basic computer science and coding skills as part of a chemistry or materials science education. Schools around the world are beginning to update their curricula to this effect, but we still need to see more incorporate this essential training. The future of the sciences is digital.

How do initiatives like Acceleration Consortium, and a conference like Accelerate, help advance the field?

We are at the dawn of truly digitizing the chemical sciences. Coordinated, joint efforts, such as the Acceleration Consortium, will play a crucial role in synchronizing efforts not only at the technical but also at the societal level, thereby enabling the worldwide implementation of an ‘updated’ version of chemical engineering with unprecedented advantages for humanity at large. The consortium also serves to connect academia and industry, two worlds that could benefit from a closer relationship. Visionaries in the commercial sector can dream up opportunities, and the consortium will be there to help make the science work. The groundbreaking nature of AI is that it can be applied to any sector. AI is on a trajectory to have an even greater impact than the advent of computers.

Accelerate, the consortium’s first annual conference, was a great rallying event for the community and was a reminder that remarkable things can come from a gathering of bright minds. While Zoom has done a lot for us during the pandemic, it cannot easily replicate the excitement and enthusiasm often cultivated at an in-person conference and which are needed to direct research and encourage a group to pursue a complex goal.

Anatole von Lilienfeld at the first annual Accelerate conference (photo by Clifton Li)

What area of chemical space fascinates you the most?

Catalysts, which enable a certain chemical reaction to occur but remain unchanged in the process. A century ago, Haber and Bosch developed a catalytic process that would allow the transformation of nitrogen—the dominant substance in the air we breathe—into ammonia. Ammonia is a crucial starting material for chemical industries, but also for fertilizers. It made the mass production of fertilizers possible and saved millions of people from starvation. Major fractions of humanity would not exist right now if it were not for this catalyst.

From a physics point of view, what defines and controls catalyst activity and components are fascinating questions. They might also be critical for helping us address some of our most pressing challenges. If we were to find a catalyst that could use sunlight to turn nitrogen rapidly and efficiently into ammonia, we might be able to solve our energy problem by using ammonia for fuel. You can think of the reactions that catalysts enable as ways of traveling through chemical space and to connect different states of matter.

Erin Warner

Researchers use AI-powered database to design potential cancer drug in 30 days

Read more about the study in the Toronto Star

'No small feat': U of T's Anatole von Lilienfeld is using AI to explore the vastness of 'chemical space'