GFlow Neural Networks to help discover new drug molecules and generate candidates.

Recently, a discussion between several top AI minds of the world and Vietnam: Professor Yoshua Bengio (Mila Quebec AI Institute), professor Ho Tu Bao (VIASM), Dr. Truong Gia Binh and Dr. Phong Nguyen (FPT Corporation), explored the usage of machine learning to transform the way we make new drugs and approached ethical AI.

HANOI, Vietnam, November 28, 2022 – The demand to find new ways to combat disease is rising by the day. Humanity is always at risk of a new pandemic, and the mutation of viruses creates resistance to antibiotics. According to experts, it has caused both high fatalities and economic value losses. “There’s already 1.2 million deaths per year, and it’s going to grow to 10 million deaths per year,” said Professor Yoshua Bengio (Mila Quebec AI Institute). “Economic costs are also rising, and it’s projected to be 100 trillion US dollars by 2050.

To combat this, prof. Bengio has been looking into utilizing Generative Flow Networks, or GFlow Nets – his team’s ML technique for generating compositional objects at a frequency proportional to the associated reward – to discover new drug molecules and generate candidates. These findings were published in 3 recent papers at renowned AI conferences.

According to prof. Bengio, one of the greatest areas of growth is at the intersection of AI and biotechnology for the next decade, thanks to its ability to reprogram the DNA of organisms and synthesize new drug molecules.

However, the number of potential new drugs is vast, and it takes decades to test which drug would work in treating which subject. Here, ML can be used to represent sample candidate experiments and shorten the time to give an educated guess as to “what chance a candidate drug is going to do the job”. But questions arise regarding ML ability to acknowledge its limitations and uncertainty, and to create diversity in candidates.

“We would like our ML system to generate candidates to be different from each other,” said prof. Bengio. “It’s important as if one drug candidate does not work, “we still have other candidates that are quite different. At the end of the day, we have a greater chance of having a drug that’s going to work.”

To decide which candidates to zoom in on since the number would be too large, prof. Bengio and his team propose using generative models, benefiting from neural nets’ ability to imagine – usually for synthesizing new images. In this case, instead of images, neural nets can synthesize molecules, and through training can be used in scientific experiments. This particular ability opens many opportunities, as it “can be applied to any kind of objects we want to experiment over, molecules, materials, even potentially, software.”

Stemming from these ideas, prof. Bengio and his team dug deeper using their GFlow Nets. In his findings, GFlow Nets proved to be working efficiently in sampling with probability and combining with the notion of Bayesian uncertainty, finding a more diverse set of candidate solutions compared to existing methods. Finally, in order to “build a really good model of data that is never going to be overconfident”, prof. Bengio looked into causality: using GFlow Nets to generate all the causal graphs that are compatible with the data.

“Approaches like causal machine learning are still in their infancy,” he said, “to conceive of the learning problem not as just learning one distribution but learning a whole family of distributions, corresponding to different context, different intervention, experiments and environments.”

Still, he noted on the challenge the current ML system faces

“If we can learn a causal model well, we can generalize all of those distributions”

the professor concluded.

Given AI’s direct influence on people’s welfare, the top minds also discussed its ethical concerns. In the Q&A, prof. Bengio addressed several issues of the long-term effect of new drugs, the misuse of AI to create toxins, as well as discrimination of data provided to AI and ML. He suggested that through illustrative research, people’s awareness of the dangers can be raised. Then, communities & authorities can create new legislations, social norms, guidelines and penalties to minimize the chance it materializes.

Based on his experience in analyzing medical data research in Japan and supporting Electrical Medical Records in Vietnam, prof. Ho Tu Bao relaid the principles of ethical AI. As protecting the privacy of people’s data is the top priority, prof. Bao pointed out the First Principle of Do No Harm, the accountability and transparency in the process of making AI, which means AI makers can answer for their products.

From the organizational point of view, Dr. Truong Gia Binh shared his aim to initiate immediate actions through educational activities. He suggested raising AI literacy by bringing AI courses to everyone, from primary to university level. His second idea is to align the education system, first at the FPT Education, with 17 UN sustainable development goals. FPT has also tried to implement practical projects for the pupil to learn from realistic experience, especially in the field of social responsibilities and AI for social good.

“AI will affect their lives later, let’s give them a chance to use AI to increase their efficiency in their daily lives”

Dr. Binh finished.

During the event, FPT and Mila Quebec AI Institute also celebrated their 2-year partnership, with achievements regarding young talents development. The representatives look forward to continuing the partnership into the future.

The event was as part of Tech Innovators – the webinar series on tech trends and experts’ insights, organized by FPT Software. For many years, the series has acted as a forum for international top minds to discuss the latest technology discoveries. With engaging topics, it has attracted thousands of people tuning in on a monthly basis.