This interview is part of a series of conversations with creatives and professionals who deal, now every day, with the advantages and limits of artificial intelligence
Leading the Molecular Modeling and Drug Discovery laboratory of the Italian Institute of Technology and also responsible for Computational Sciences, one of the 4 domains into which IIT is divided, Marco De Vivo is first and foremost a chemist. And for a few years now he has been using it more computers and artificial intelligence than mixers, centrifuges, stills and test tubes to discover new substances and new drugs.
After his bachelor’s and doctorate, he passed over 5 years in the United States as a researcher in the academic world (at the University of Pennsylvania) and at Yale in the biotech sector (in Rib-X Pharmaceuticals, now Melinta Therapeutics). He returned to Italy in mid-2009, founded two drug development startups, one for cancer and one for neurodevelopmental disorders, and joined IIT to do what he does best: “I apply the science computational to understand the biological and biochemical processes” which are the basis of diseases. From there, he and his team create so-called small molecules (amino acids, lipids, sugars, fatty acids, alkaloids, and so on), then combining them together to create new drugs.
Change! Journalist Vittorio Roidi: “AIs are a useful tool, but rules are needed to use them. And they are needed now” by Emanuele Capone 22 January 2024
You create them using the computing capabilities of super computers and artificial intelligence. Don’t you do experiments anymore?
We do both, because computational sciences and human capabilities combine: we continue to do experiments and tests, which is something that cannot be ignored, but many are first simulated on the computer, so that we can focus only on those that could produce results, the most promising and most likely to be successful.
Concretely, what are the advantages of using AI in the pharmaceutical industry?
To give a practical example, it’s like designing a bridge (which is a much simpler thing than what we do) by being able to simulate whether it will stand before trying to make it stand. There are significant advantages from the point of view of time and from the point of view of costs: it avoids investing resources in experiments that would fail.
An Italian Tech Academy master’s degree to understand how to use Artificial Intelligence in professional contexts
Why are AIs so useful?
The secret to it all is machine learning (here we explained what it isndr)one more tool, one more way to analyze and understand the data available: in many different fields, from genetics to medicine, from chemistry to biology, AI are able to analyze enormous quantities of data and find connections even unthinkable for the human mind.
By connections we mean combinations of molecules that could give rise to drugs useful for humans, but why is all this happening now?
It is happening now, understood as for a few years now, because 3 ideal conditions have occurred now and all together: we have a large amount of data, so vast that people would struggle to analyze it without IT help, we have high quality data and we finally have the computing power needed to manage this information.
Is there a year that we can point to as a turning point?
The fundamental one was undoubtedly 2021, when Google and DeepMind they released AlphaFold, a huge database of around 200 million proteins in 3D: it was understood that AIs were also very good at predicting and imagining protein structures that could fit together, function and be efficient. It was an incredible step forward, which left all competitors behind. Anyone can starting from the database (which is this, ed.) to try to combine molecules, trying to simulate and understand how they work.
Where does human professionalism fit into all of this?
In my experience, machine learning will increasingly have a positive impact on science and research, especially to help understand things faster but there are at least two aspects in which people remain and will remain fundamental.
Which ones are they?
First of all, it takes time to have the quality data I was talking about before, because it needs to be standardized and inserted into databases that machines can study. And this care work must be done by people.
And what is the other aspect?
I see great and understandable optimism about the use of machine learning but human intervention, the creativity, the ability to have the right intuition at the right time, will always remain fundamental in my opinion. I repeat: machine learning is a tool useful, but it must be applied in the correct way and to the right problem.
In what sense?
I remain convinced that it is a good thing if understood and understood, but it cannot be the solution to everything and we must not exaggerate with expectations. Although it is undeniable that it has allowed us and will allow us to go beyond the limits we had and still have: we are in possession of a lot of data, it is time to use it and take advantage of it.
About this: Meta has announced that it will invest billions of dollars in the development of its new AI. That’s a lot of money. How expensive is it to take advantage of these tools?
In 2020, IIT invested just under 3 million euros to create Franklin, a cluster of 300 GPUs used to do these jobs, which helps both the institute’s researchers and some startups within IIT, such as my own IAMA Therapeutics. By doing research at an atomistic level with the use of machine learning, we arrived at clinical trial phase immensely faster and with very little experimentation. We have saved many years and a lot of money.
But is all this computing power really necessary?
Yes, without a doubt: according to estimates, there would be around 10 to 60 possible combinations of molecules (for context: 10 to the 12 is 1 trillion, ed.) and an AI is able to explore all these possibilities and navigate the so-called Chemical Space at a speed unthinkable for a human being. And when he finds something useful to solve the problem in front of him, the researcher can then do an experiment and verify.
Change! The photojournalist Fabio Bucciarelli: “As long as a person runs away in Gaza, AI won’t scare me” by Pier Luigi Pisa 15 January 2024
And the researcher? How will this role change in the future?
In the world of work there will undoubtedly be a change, a shift towards data sciencestowards roles such as data scientist or data warden, fundamental for having the quality data mentioned above and for which there is already a lot of demand even from younger people.
But don’t we risk seeing this profession eliminated?
I think not, because in any case we cannot ignore experimental science and experiments. Let’s hope there will be research robots but there will always be someone who will have to teach these robots how to act, how to work, how to do research. AIs are perfect for quickly understanding which experiments to prioritize, which to do and which not to, but just as a truly good suit is made by the tailor, a truly quality drug is made by human sensitivity.
@capoema