Video caption: This conference on biochemical modeling is interesting in several respects
Can AI improve medicine? From those involved in clinical processes, the answer we hear is a resounding “yes!” But how will this be done?
We see that by properly leveraging data, AI systems can multiply productivity and bring new models to medical practices.
Moving on to some of our early discussions about how to use AI to improve treatments, we hear more about biochemical modeling and related concepts from Peter Mikhael, who describes some of these next-generation initiatives that will almost certainly have a major impact on hospitals. systems in the years to come.
First of all, Mikhael covers three main areas: risk and outcome prediction, protein modeling, and chemical structure modeling.
“Chemistry is a very broad and diverse field,” he says. “And yet, in this big space, we would like to find the one thing that could potentially be therapeutic.”
Regarding this first prediction goal, Mikhael talks about building models to know the past results of data sets and reminds us that chemistry exists in a context.
“The context is usually that of a biological system,” he says. “So we need to think about biomolecules: things like proteins and peptides. We need to develop approaches to model these things.
Mikhael talks about early detection tools that will make cancer screening easier. In terms of modeling chemical structures, he gives the example of finding the best antibiotic (which was also covered in this CSAIL talk by David Sontag).
It also addresses certain aspects related to molecular properties: representing a molecule graphically and testing it on a computer before testing the elements in vitro.
“We do consistency checks,” he says. “You don’t want something that’s toxic to the human body, that’s toxic to bacteria… of course, you also want it to be a structurally new molecule, because if it’s something that already exists, you could rediscover wheel. .”
As for proteins, he talks about mutations, diseases and protein-ligand interactions:
“Normally you don’t have…any abnormalities…as soon as you have a few mutations, potentially, now you have a disease,” he says. “In this case, for this particular protein, you have uncontrolled cell growth. And that’s something you’d like to sort of address. And the molecules that we talked about, these drugs, have to have a particular shape, they have to have a particular conformation that fits in the pocket, so that it inhibits whatever the potential protein does, right? This is the idea behind a lot of potential drugs. So how can we do this? How do we actually model this thing, this interaction between the protein and the little ligand, such that we know whether or not we have a good drug?
Next, he delves deeper into the idea of torsional diffusion for molecular docking, something else he says students in the lab have been working hard on. Listen to this part:
“Let’s say you have a protein, in this case, some sort of SARS-COV-II spike protein, and then a small molecule, and you want to be able to tell exactly where in the big space that small molecule is located. protein. he explains. “So what you take… you take the same molecule here, just shown in different random places around the protein, and you want to learn how to iteratively move this protein, twist it, reshape it, such that ‘she’s adjusting.’ in the correct pocket where it should be, and you train a diffusion model that does this: changes all the twist angles, moves it, and you can, over time, learn to just place a molecule somewhere around the protein, and we will find the right pocket for it.
This approach, he says, is something students call “DiffDock.”
“(It’s) the first deep learning method that’s better than physics-based methods, which, as you would imagine, have been, you know, this kind of status quo for a long time.”
This all fits nicely into some biochemical research that’s getting a lot of attention right now and is consistent with a lot of what we do, both in academia and in business. Think, for just one example, of the types of things covered by Ark Invest funds, in genomics, AI/ML, etc. Molecular modeling and some other content here is part of the mix, so perhaps in addition to data scientists, medical professionals, etc. Many business people should watch this video.
Portrait of Peter Mikhaël
