Marrying data and digital
Surprise, surprise
content-image
Science
00

A smart ally in the fight against malaria

At the turn of the millennium, when Novartis started its Malaria Initiative to help rein in the disease in Africa and Asia, artificial intelligence was a distant dream. Two decades later, the technology could prove instrumental in the fight against the disease, which still kills more than 400 000 people every year, most of them children.

Text by Goran Mijuk, Photos by Janet Delaney

scroll-down
Home
en
de
zh
jp
Share
Share icon
content-image
Enter fullscreen

“While we have two compounds in clinical trials at this stage, I wanted to examine, as part of a proof-of-principle study, whether William’s system could be useful to produce a lead compound and explore whether AI could help us in the fight against malaria going forward.” - Armand Guiguemde

arrow-rightMarrying data and digital
arrow-rightSurprise, surprise

Published on 17/03/2022

Armand Guiguemde, a scientist at the Novartis Institute for Tropical Diseases (NITD), is skeptical by nature, especially when it comes to his work as a researcher in the malaria field, which faces the perennial challenge to produce new compounds to keep the disease in check.

This is how I got to know Guiguemde in late 2020. We were discussing a quote he was initially reluctant to share, when our chat shifted towards artificial intelligence (AI) and he told me about a project he was working on together with data scientist William Godinez from the Novartis Institutes for BioMedical Research (NIBR).

A coding wizard, who has launched a number of research projects since he joined Novartis five years ago when the company started to boost its digital efforts, Godinez had developed an AI program back in the summer of 2019. The goal was not only to provide researchers with novel and high-quality ideas for compounds, but to do so in record time – thus helping to shorten the typical timelines and costs of drug discovery projects.

Godinez’s project had caught Guiguemde’s eye during one of the regular scientist get-togethers in Emeryville, where they both work, in autumn 2019. “It was during one of our poster sessions when I learned about William’s program,” Guiguemde remembered. “I was interested to learn more and test it.”

Fighting resistance

Guiguemde’s curiosity was driven mainly by the complexity of malaria science and the need to continually develop new compounds. Because malaria is a mosquito-borne disease caused by a parasite, existing medicines lose their power over time as the parasite builds resistance to treatments.

This phenomenon has happened repeatedly throughout history. An extract from the bark of the cinchona tree had been the standard malaria therapy for almost 300 years before quinine was used in the early 19th century. Later, quinine was replaced by chloroquine. But, like subsequent drugs such as mefloquine and atovaquone, all of these compounds lost their efficacy.

In fact, in the late 1990s, the World Health Organization (WHO) feared that some parts of Africa and Asia could become uninhabitable as the disease killed more than 1 million people every year as standard therapies had lost their potency.

This is where Novartis came in. It had developed anti-malarial Coartem® at the time, joined forces with the WHO to set up the Malaria Initiative and created the Novartis Institute for Tropical Diseases to develop new compounds in preparation for future resistance.

This investment has paid off handsomely from a global health perspective. Since 2001, more than 1 billion Coartem packs have reached patients in endemic countries, helping rein in the disease and supporting the United Nations in its effort to achieve one of its Millennium Goals. Furthermore, the NITD, together with several partners, is currently testing three new antimalarial compounds in the clinic – the fruit of 20 years of continual research.

content-image
Enter fullscreen

“AI can achieve a lot of great things from a computational point of view. However, if the program lacks clean and structured data points, results can be disappointing.” - William J. Godinez

Mar­ry­ing data and di­gi­tal

“While we have two compounds in clinical trials at this stage, I wanted to examine, as part of a proof-of-principle study, whether William’s system could be useful to produce a lead compound and explore whether AI could help us in the fight against malaria going forward,” Guiguemde said, adding that his initial hopes were not particularly high.

Guiguemde’s mixture of curiosity and skepticism was well grounded. Artificial intelligence has been on a stellar rise over the past decade and has left its mark on almost every industry, from social media to finance. But breakthroughs in the realm of drug research and development have been relatively few so far.

This is down to the particular nature of AI. In order for an artificial intelligence program to produce meaningful results it needs to be trained with massive amounts of data – something which is hard to come by, especially in the area of drug research and development.

On top of that, data also needs to be structured so the computer can read it, which in many cases is time-intensive and costly, especially in healthcare. “AI can achieve a lot of great things from a computational point of view. However, if the program lacks clean and structured data points, results can be disappointing,” Godinez explained. “The power of AI only comes to fruition when we marry it with clean and structured data.”

A German drug hunter

While fully aware of the challenges that come with AI, Godinez is convinced the technology will help the healthcare industry in the long term. Having worked in various units at the Novartis Institutes for BioMedical Research, he has brought AI skills into traditional drug research domains such as cellular image analysis and bacteriology, where AI today is increasingly becoming a core research component.

But he has bolder aspirations, especially when it comes to producing drug molecules. When he heard about an open-source program developed at the Massachusetts Institute of Technology intended to automate the design of drug candidates, Godinez jump-started the project together with his colleague Eric Ma, who refined the program to fit it to the needs of Novartis.

Their coding efforts resulted in a program called JAEGER (short for the rather unwieldy Junction-Tree Variational Auto-Encoder GenERative Modeling), which means hunter in German – in homage to Godinez’s student years at the University of Heidelberg as well as to all drug hunting activities at Novartis. “It didn’t take us a long time to adapt the program,” Godinez said. “But what we really lacked was data, and so we decided to showcase our project during one of our poster sessions. We were more than happy when Armand showed an interest in working with us and was ready to share his data.”

And what a treasure trove Guiguemde had. “We have massive data from more than 60 000 molecules in malaria research going back some 20 years, when the Novartis Institute for Tropical Diseases was created,” Guiguemde said. “So, why not use it and see what comes of it.”

content-image
Enter fullscreen
Sur­pri­se, sur­pri­se

Guiguemde’s “let’s do it” attitude and willingness to collaborate with his colleagues proved to be spot-on as JAEGER came up with a result that surprised everybody. The drug hunting program not only generated molecules within a few hours, the result also convinced chemists that the AI-designed compounds were looking promising and could potentially be synthesized.

This result was far from expected. Often, AI programs – if they lack the right data input and algorithms – sputter out useless results. Either the suggested molecules are too complex to synthesize or too toxic for human use. In some cases, AI-designed molecules can simply be nonsensical.

“To be honest, I was amazed by the result,” Guiguemde acknowledged. “JAEGER was able to propose high-quality molecules very quickly; usually, it takes months if not years before we hit upon a highly promising compound. Synthesis and experimental validation at the NITD labs also showed that the compounds had excellent antimalarial activity, making our proof-of-principle study a complete success.”

Prospect of AI

Godinez and Armand, who have since published their findings in Nature Machine Intelligence and have also been reviewed in Nature's News and Views section, are convinced that AI can do more in future. “With our JAEGER program, we were able to show in principle that AI can produce chemically valid structures and produce novel chemical matter with desired physicochemical and bioactivity properties. This will help our drug discovery efforts going forward,” the two explained.

To that end, Godinez is also participating in a Novartis-wide collaboration with Microsoft to deploy these technologies across the company. The cooperation between the IT and pharma giant, which was announced in 2019, is a multi-year alliance which will leverage data and AI to transform how medicines are discovered, developed and commercialized.

But AI could do even more, Godinez is convinced: “Today, many of our drug research and development processes are already informed by AI tools that help us make decisions quicker. I can imagine bringing these systems together, which could help us not only develop compounds faster but do the ensuing development steps in record time.”

However, this will remain an uphill struggle and will not replace classic science and traditional data gathering, Godinez explains. “AI, at its best, will only work if we can train it on clean and structured data and continue to do the science we have always done. However, we can hopefully do it much faster than in the past with the help of AI.”

In the face of more than 400 000 malaria deaths every year, most of them children, nothing could have more urgency.

icon

Home
en
de
zh
jp
Share
Share icon