A Robot Chemist

Lead Research Organisation: University of Manchester
Department Name: Computer Science

Abstract

Eve is an artificially-intelligent 'Robot Scientist' designed to make drug discovery faster and much cheaper. She has already discovered that a compound used in soap and toothpaste might also be used in the fight against drug-resistant malaria, demonstrating its success. The proposal is to now give Eve the ability to do chemical reactions and to synthesise new compounds.

Eve inhabits an enclosure 2.5 meters in length, 2 metres wide, and 93 meters high. It consists of two robot arms, surrounded by equipment regularly found in laboratories for dispensing liquids into a large number of wells lined up on plastic plates, then incubating and testing them. But by integrating together instruments usually separated into different departments, Eve can do tests and interpret the results, and go on and use that knowledge in further tests faster. We will now give Eve the power to design and make her own, new compounds before testing their potential for drug discovery.

For the majority of medicines available today, scientists view drug molecules as nanometre-scale keys that slot into similarly sized protein or enzyme locks in cells in our bodies. Drug screening tests put these locks using biological systems that trigger a signal, such as a fluorescent flash, when a molecule fits into it like a key.

While pharmaceutical industry screening can identify positive signals known as hits, Eve is also independently able to follow up and check if the hits were true prospects, known as leads. But simply screening and following up hits is not where Eve's greatest promise for drug discovery lies. Instead, by learning from the results from those tests, Eve is able to do what it currently takes teams of chemists and biologists many months to hammer out. Drug researchers currently already use software that employs 'machine learning' to take screening results and create a 'quantitative structure-activity relationship'. This is a mathematical function that relates the composition, shape and properties like fattiness and electrical change of the molecules, to how good drugs they are likely to be. Using such models scientists choose which molecule to make and test next.

Currently, Eve can only learn to predict which out of a large set of ~15,000 compounds would be hits. The proposal is to add to Eve the ability to also synthesise novel compounds. In particular, we will program Eve to be able to carry out a chemical process known as 'late stage functionalization'. This is the introduction of a medicinally-relevant chemical group to existing drug-like molecules in Eve's library. This will enable Eve to make new chemical entities and to form an extended collection of drug-like molecules. We will program Eve to use machine learning to (1) Learn how to best to design drugs using late stage functionalization, and (2) learn which molecules are most likely to undergo successful late stage functionalization.

An important goal of our project, therefore, is for Eve to develop, optimize and 'road test' a new and important chemical process that will be of great use to molecule-makers around the world. However, the main project goal is to make drug discovery cheaper and faster. This will enable the development of treatments for diseases currently neglected for economic reasons, such as tropical and orphan diseases, and generally increase the supply of new drugs, and so potentially improve the lives of millions of people worldwide.

Planned Impact

Impact Summary
The proposal is an ambitious one with a high potential for significant technological, medical, economical, and societal impact.


Recognised National Importance - AI
The importance of AI research to the UK is recognised at the highest level: On 22.11.17 The Chancellor announced in his Autumn Budget £75M for AI research and development; On 27.11.17 The UK government's "Industrial Strategy: building a Britain fit for the future" white paper was published. The goal is to "propel Britain to global leadership of the industries of the future - from artificial intelligence and big data to clean energy and self-driving vehicles." Prime Minister.

Recognised National Importance - Chemistry
As molecules are prepared and manipulated in the day-to-day work of academic scientists and scientists in established (pharmaceutical, agrochemical, contract research organisations) and emerging (organic electronics, biotech /biopharmaceutical) industries in the UK, our studies aiming to develop the first Robert Chemist are aligned with the needs of UK industry (and academia).


Technology Readiness
The proposal aims to take the concept of a Robot Chemist from TR2 (Technology concept and/or application formulated) to TR4 (Early proof of concept demonstrated in the lab).


Commercial Opportunities
The integration of AI with Drug Design is currently a "hot topic". RDK has close ties with Exscientia, with whom he collaborated on a previous EPSRC grant. Since then the Dundee based start-up have done numerous large deals with big Pharma. The most notable of these is a deal worth potentially Euro 250M with Sanofi. The leading UK company in the area of AI and drug design is Benevolent AI. They are a 'Unicorn' company now worth £2Billion.

In addition to a number of successful academic collaborations, DJP has active funded collaborations with industrial teams (AstraZeneca x 2, Lilly UK). DJP has received 30 grants from UK Industry (AstraZeneca, Lilly UK, GlaxoSmithKline, Syngenta, Pfizer, Avecia, Novartis, Celltech, MSD, OSI Pharmaceuticals, SAFC-Hitech, Pentagon Chemicals Ltd), received Pfizer and AstraZeneca Strategic Funding, and has consulted for three companies (Antabio, Syntor Fine Chemicals, PZ Cussons). Knowledge exchange in the PI's team has also been facilitated through the delivery of invited Industrial courses for AstraZeneca (2006, 2010) and GSK (2009, 2013-2016). Knowledge Transfer Partnerships (ACAL Energy, Pentagon Fine Chemicals Ltd) have successfully implanted DJP's synthetic expertise into two companies. Finally, medicinal chemistry teams at AstraZeneca (UK) and Theravance Inc. (biopharmaceutical company, California USA) have used DJP's processes to generate lead compounds during industrial projects in important therapeutic areas.

Follow-on Funding
The proposal is for an initial 'feasibility study'. To continue the research we will require follow-on funding. We will investigate opportunities for both grant and commercial funding. Potential sources of grant funding are the EPSRC (especially a possible 'Digital Chemistry' call - see the recent EPSRC workshop), H2020, the Welcome Trust, DARPA, etc. Potential commercial sources of funding are large Pharmaceutical and Health product companies, and new AI/Chemistry companies that are starting to appear, e.g. Exscientia, Benevolent AI, etc. We will also approach our VC contacts, as it is our experience that they are willing to invest in such early stage developments.


The evidential support to produce a high quality follow on funding application will come from two main sources: publications, and Intellectual Property (IP)

Research capacity building
The project will train two PDRAs in key areas of future science and technology: AI, machine learning, laboratory automation, synthetic organic chemistry.

Publications

10 25 50
publication icon
Coutant A (2019) Closed-loop cycles of experiment design, execution, and learning accelerate systems biology model development in yeast. in Proceedings of the National Academy of Sciences of the United States of America

 
Title Closed-loop AI experimentation 
Description Our work is now causing a revolution in materials science. 
Type Of Material Improvements to research infrastructure 
Year Produced 2009 
Provided To Others? Yes  
Impact Closed-loop AI experimentation