Over the past few decades, advancements in computing power, machine learning and data processing have changed the world; now they represent one of the biggest promises for the future of medicine… Among its numerous medical applications, computer technology allows for building complex computational models – of a protein or an organ, for instance – to run simulations and test the possible outcomes of a new therapy or drug. Coupled with our knowledge of the human genome, these predictive tools constitute the basis of personalized medicine, an approach tailoring treatments to each individual patient.

Riccardo Sabatini, scientist and entrepreneur, is one of the founders of Orionis Biosciences, a U.S. biotechnology firm working on genome-based drug design. A renowned international speaker (his TED2016 talk received over 1.7 million views), Sabatini is specialized in numerical modeling of complex systems, which over the years has led him to work – along with scientists of the caliber of Craig Venter – in market forecasting, artificial intelligence and computational genomics. Sabatini was also one of the young innovators in health invited by Barack Obama to speak at The White House Frontiers Conference 2016.

We asked Sabatini about his current projects, the state of computational medicine and the impact of supercomputers.

Q: what are you working on at the moment?

A: I’m working on a new platform to discover drug interactions across the entire human proteome, with an extension to hundreds of rare-diseases alleles. Drug design went through several phases in history, moving from a phenotypic screening process to a target-based one. I believe genome-wide drug design represents the next era of medicine, allowing us to tackle diseases we’re still struggling to hit.

Q: How close are we to computer-based, personalized medicine?

A: It’s a very complex challenge, and still far from everyday applications, but we’re getting better and better. Right now, we are focusing on how to predict if a patient would respond in an unexpected way to a specific drug – an indication that can help clinical teams to select another molecule or tune the dosage of the current one. Or to discover new uses for old drugs. Thanks to modern crystallography and new numerical methods, we’ll soon be able to run these tests in silico (using computer software or simulation).

Q: How does computational, genome-based drug design work?

A: Starting from a three-dimensional “picture” of a protein, a crystallographic reconstruction typically realized with nuclear magnetic resonance and electron cryomicroscopy, we can simulate the possible folds of the protein, searching for possible “attack” points for a drug. Also, once the protein-drug interaction is known we can analyze mutations of that protein, specific for a certain subpopulation or single individual, simulating the different behavior and computing the reduction or growth of the drug engagement – a representation of the drug potency.

Q: What is AI’s role in computational drug design?

A: Designing a drug for a specific site of attack in a protein – a pocket – is incredibly hard, and machine-assisted design is starting to be of enormous help in the chemical-space exploration. As an analogy, imagine you have to build a key for a lock you don’t know in detail – an experienced locksmith can possibly take days or weeks of trial and error, while if you could ask a supercomputer to create and try millions of keys and test them on the numerical model of the lock, you could speed up this process by orders of magnitude.

Q: How are supercomputers helping?

A: To compute, and then predict, if a drug will engage a pocket of a protein correctly or not, you first have to solve an incredibly hard numerical problem: the quantum mechanical interactions of all the atoms in the protein – tens of thousands – with the drug. At this scale only a large supercomputer has enough memory and CPU power to solve one of these configurations in a reasonable time and iterate it on the thousands of drugs the research team wants to try.

Q: What are the main challenges of computational drug design?

A: There are still many mathematical challenges to solve. The bare simulation of all atoms is often infeasible even for the largest supercomputer, and we need smarter methods to reach a solution more quickly. Protein-folding algorithms using meta-dynamics as well as machine learning-driven solutions or evolutions of these concepts are very exciting, as they help us save trillions of computations and scale faster to more and more targets and proteins.

Q: In a recent interview with Forbes, you mentioned the importance of customized drug design in the fight against cancer. Is this something you are currently working on, and could you tell us a little more about it?

A: Oncogenes, a term referring to the mutated proteins helping cancer to survive and spread, are fundamentally important targets to “attack” with small molecules. The more precision we get in cancer sequencing and in drug-mutations matching, the better we’ll be able to deliver precise therapeutics. It’s a long process, and I promise to keep you updated!