Daphne Koller does not mind hard work. In 1995 she moved to the computer science department of Stanford University, where she spent the next 18 years full-time, before co-founding the online education giant Coursera. She spent the next four years there and remained co-chair until last month. Koller spent less than two years at Alphabet's longevity lab, Calico, as the first Chief Computing Officer.
Koller was reminded of her passion for using machine learning to improve human health. She has also been reminded of what she dislikes, which is a futile effort, something that has been plaguing the drug development industry ̵
In Fairness These calculation methods have also gotten a lot better lately. No wonder Koller seized the opportunity to start another business last year, a drug development startup called Insitro, which has since raised $ 100 million in Series A, including GV, Andreessen Horowitz and Bezos Expeditions , As can be noted, the company recently partnered with Gilead Sciences to find drugs to treat liver disease called nonalcoholic steatohepatitis (NASH), which is due to all the human data on the disease that Gilead has amassed over the years Has.
Later, insitro might even target larger epidemics, including possibly Alzheimer's disease or Type 2 diabetes. Surely there is reason to be optimistic about what can be achieved with it. Koller said a few days ago to a group of enthusiastic participants at an event hosted by the editor: "We are now in a moment of history where a confluence of technologies has emerged that make it really big, interesting and pathogenic To achieve results. relevant records are produced in biology. In parallel, we see. , , Machine learning technologies that are able to understand this data and provide new insights that can hopefully cure disease. "
It all sounds like we've heard it in the last few years, but when you come from Koller, you get the sense that, despite the secrets of human biology, we're finally approaching. Here are some excerpts from Koller's interview with the journalist Sarah McBride of Bloomberg. You can also follow the conversation below.
Why Insitro has partnered with Gilead (and, moreover, this could prove lucrative as milestones of up to $ 1 billion have been set for the successful development of targets for NASH):  There are quite broad categories for which our technology is well suited. We are really interested in creating something like "disease in the shell" models – places where diseases are complex, where we really did not have a good model system and where only typical animal models were used [for years, including testing on mice] are not very effective – and the creation of these in vitro models to generate very large amounts of data that can be interpreted with machine learning.
There are a number of diseases that are suitable for this type of approach. NASH was one of them, in part it was the suitability of our technology for this disease, and in part Gilead was just a really good partner for it because they have a whole slew of human data from some of the clinical trials that exist [which give us] Access to two complementary data sources. One is what happens to the disease in large human cohorts, and one is what happens when you look at what the disease does in the shell in vitro, and then see if we see what we see in the shell Using Machine Learning to Predict What Happens We see in humans.
How Insitro Dates Differently Than Big Pharmaceutical Companies:
Pharmaceutical companies say, "We have lots of data." And you say, "What types of data do you have? "And it turns out they have datasets that are each stored in a separate table on someone else's laptop. There are metadata that have not even been recorded yet. For them, it's like, "Yeah, I did the experiment and obviously I recorded what I needed because there's no point throwing it away", but they do not see it as something you build a business of.
We come in a different way. We say, "This is the problem you want to solve, if only we had a model that could tell us the result of this experiment without having to do the experiment because it is costly or complicated or even impossible [because it would involve perturbing a living human’s gene]. "Well, machine learning has become really good at creating predictive models when you give it. It's the right data to train the model, so we're able to access data just for the purpose of training machine learning models We think of [these models] as small crystal balls that you can use to avoid [these more expensive or complicated] experiments.
About the Impact of the National Institutes of Health's All of Us Research Program is an attempt to gather data from one million or more people living in the US to accelerate research and improve health, in part through individual differences to be recorded in Lifestyle, Environment and Biology:
I would say, if any, that the US is a bit late in this case. Several countries have already formed a number of national cohorts. The two countries that are currently the most developed are in Iceland and the United Kingdom, but there is also one in Finland and one in Ireland and even in Estonia, where they have taken a large population from that country and measured their genetics, but also have measured a number of traits about these people, including blood biomarkers and urinary biomarkers, as well as behavioral aspects and physical aspects and imaging. What you have now (in these countries) is a record that says, "Nature has disturbed this gene" and "We see this effect on humans."
[In the UK, specifically, where they started their program five years ago and recruited 500,000 volunteers who agreed to physical and cognitive and blood pressure testing and images of the brain and the abdomen, among other things] It's an unbelievably extensive record [from which] Discoveries are noticeable almost weekly.
… This is not only important for gene therapies, but also to identify targets that really make a difference, as most drugs used in clinical trials fail. And with most I mean 95 percent. And most medications fail because they target the wrong things. They target proteins or genes that do not affect the disease they are supposed to affect. The recent, very obvious failures of Alzheimer's drug trials – actually several in a row – were almost certainly due to the fact that the protein they target is amyloid beta, not the right causative factor for the disease.
What researchers can now do with stem cells, which would have been impossible just a few years ago:
[There are now] Tools that produce not only large amounts of data but also large amounts of biologically relevant data could. So we did experiments with cancer cell lines. , , but it is not a very disease relevant model. Today, we can take a small sample of skin cells and use the so-called Yamanaka Factor to reprogram these cells to stem cell status. These are the cells that effectively exist in the womb. And these cells are able to differentiate into neural cells or liver cells or heart cells, and these are very disease relevant because they represent human biology. You can now take these cells from patients and healthy people and determine if there are differences in their presentation. Readers, we could show more of the transcript here, but we strongly recommend that you pursue the conversation with Koller.
If you use this text as a starting point, you want to start listening from the 13-minute mark. It's definitely worth listening to what she has to say, including cystic fibrosis, spinal cord dystrophy in infants, and why the "mouse models" we've long relied on for a variety of apparently ubiquitous diseases "from bad to bad." Hard enough, really, very bad. "I hope you like it.