Treating diseases with AI.

I’m the CEO of OccamzRazor, a startup that uses machine learning to find treatments for Parkinson’s disease and other brain ageing diseases.

In my previous career as a scientist, I became the first to identify the stem cells that form coronary arteries. This discovery was recognized as one of the top five most important breakthroughs in 2015 by the American Vascular Biology Society. I published that finding as part of my PhD in Stem Cell Biology and Regenerative Medicine at Stanford University and  was able to complete my graduate studies in a record time of two and half years. 

My work has been profiled in magazines and newspapers such as The Guardian, BBC, WIRED, and La Repubblica. In 2017, I made the Forbes 30 under 30 list in the science category, and in 2020 I was featured in the MIT Technology review list of the top 35 innovators under 35.

When the Human Genome Project was completed in 2000, researchers predicted that we would finally be able to find treatments for diseases like Parkinson’s and Alzheimer’s. Twenty years later, we’re still at a loss about how to tackle these diseases.

The reality of these diseases is far more complex and far too difficult for humans to understand at a fundamental level. Furthermore, in medicine, thousands of papers are published on a daily basis. No researcher can possibly keep up with this deluge of information. As a result, most drug trials are poorly conceived and designed. In fact, despite costing billions of dollars, the majority fails.

To solve this conundrum, we need machine intelligence: a program that understands scientific language, synthesizes medical literature, and suggests unbiased hypotheses. That’s what our algorithms do. As we start partnering with pharmaceutical companies, our AI will accelerate drug discovery and create new, life-saving drugs.

Parkinson’s is an extremely complex disease. Technically, the disease is a result of the loss of brain cells that produce the neurotransmitter dopamine, but what causes those cells to die is unclear. Different patients can develop different symptoms. Some develop tremors, some lose their sense of smell, some develop rigidity and balance problems. One thing we do know is that the number one factor that contributes to Parkinson’s is age. Although Parkinson’s only affects between 1 to 2 percent of the population aged 80 years or older, the way this disease and ageing affects our neurons and our body is very similar.

Because of this biological connection, we hope that by finding cures for this disease, we will also be finding drugs and treatments for other brain-aging disorders that will help people live longer lives in full possession of their physical and cognitive capacities. Parkinson’s, in a sense, is our trojan horse to conquer a much bigger goal: ageing itself. By doing so, we won’t only be helping the 10 million people worldwide who suffer from this terrible neurodegenerative disease, but also the billions that suffer from all age-related diseases.

To find treatments for brain diseases like Parkinson’s we have to understand its biological complexity in full: how genes and proteins and metabolites are connected and how they affect each other. This network of interactions is too complex for humans to grasp. We need machine intelligence to tackle it. We’ve invented machine learning algorithms that can read biomedical reports and precisely map out the network of connections between the different biological elements. When we analyzed the more than 20 million scientific papers that have ever been published about Parkinson’s disease, the result was the first ever holistic map of the disease, which we called the Parkinsome. This map is constantly updated. As soon as a new paper gets published or a new patent comes out, its findings get immediately ingested into our natural language processing system and included in the Parkinsome. This system can also predict biological connections currently unknown to researchers and suggest new hypotheses.  Researchers can query the algorithm: For instance, based on all the protein and protein interactions that you have seen, what molecular targets should we be looking for? This will make the process of drug discovery much faster and efficient. 

Today, more than 10 million people worldwide live with Parkinson's Disease. 

One of the first insights we gained from our work was the realization that Parkinson’s isn’t one disease, but a group of different ones, each with its distinct set of causes and symptoms.

Our primary goal is to develop cures that are individualized to the Parkinson’s patient. To achieve this, we have established partnerships with the Michael J. Fox Foundation and the Parkinson’s Institute and Clinical Center, who provided us with 25 years of clinical data that included everything from the patients' history, every medication that's ever been taken, secondary diagnoses, and age of onset.

We are now validating the AI predicted drug targets in the laboratory. For that we use skin cells from patients, convert them into stem cells and then into brain cells. We then test different treatments on these neurons, hoping to convert them into healthy cells and will apply for FDA approval to run clinical trials. We are starting to develop similar lab assays for other brain-aging disorders and will be able to discover and develop very targeted approaches to treating each disease and subgroup of patients.

As we’re witnessing with the current pandemic, the elderly are usually left behind when it comes to healthcare. This will become even more urgent in the future, as the global population ages and the prevalence of neurodegenerative diseases increases.Biomedical research is also failing: currently, the probability of success for a new drug entering clinic trials is one in ten, for brain-aging disorders, all trials have failed other than ones only treating symptoms. The pharmaceutical industry has failed to deal with complex diseases like Parkinson’s that predominantly affect our ageing populations. We want to change that by supercharging research with AI.

When I started studying Parkinson’s, I realized, immediately, it was still a poorly understood disease. To make progress would involve interconnecting information from subfields of expertise that don’t communicate: genomics, proteomics, metabolomics, etc. We needed a complete picture of Parkinson’s — a map that contained all the existing information and depicted the complex biological network of relationships between genes, proteins and all the other elements. 

To do that, we needed technology that could read and understand biomedical papers, so we started by building a dataset. My team spent days labelling sentences in scientific publications. For each sentence, we would label protein, genes, organelles, etc, and their relationships. We took that dataset to the Stanford AI Lab and created a partnership to develope a natural language processing algorithm that could read a text and identify the type of genes, proteins, organelles, the metabolites, cell types and their biological relationships. This was our first breakthrough in machine learning that paved the way for us to develop an AI that can take all biological information, from text based documents to structured genomics datasets, to map out the disease and predict curative treatments addressing the multiple causes of the disease.

In 2016, I received a call from my mother with some very bad news.

My mother lives in Germany and, unfortunately we don’t get to see each other as often as I wish, so long-distance phone calls are the only we keep in touch. 

At the time, she was feeling that something was not quite right with her. For instance, she felt that she couldn’t control her hands anymore. She had googled the symptoms and that’s what led her to go to the doctor, who confirmed her suspicions. She was diagnosed with Parkinson’s. 

I was devastated. For two days, I stayed in bed and cried. Then I made a resolution: I was going to find a cure for Parkinson’s. Finding a cure from this terrible brain-aging disease that affects my mother and so many other patients is my constant motivation.

The first time I was involved with Parkinson’s research was during my PhD. I was trying to convert skin cells into stem cells and then into dopamine-producing neurons that we could transplant into lab mice. One of the samples actually came from Google’s founder Sergey Brin and his mother. They both had a mutation associated with the disease. 

Brin was also contributing millions to Parkinson’s research. He wasn’t interested in academic research for the sake of it, he wanted to get therapies to the clinic as quickly as possible. He wanted to move fast and cure diseases.

Years later, when I started OccamzRazor, that experience pretty much influenced my philosophy. Whereas Google’s mission was to organize the world's information and make it universally accessible and useful, our mission was to organize the world’s information about Parkinson’s and make it useful. 

We’ve recently become the first to map all the scientific information about this disease. Now, we’re starting to pinpoint the cellular dysfunctions that lead to it and soon we will start to develop treatments.  

This achievement was possible thanks to the team behind OccamzRazor: a group of Ph.D.s in machine learning and neurosciences and a board consisting of Nobel laureates, experienced pharmaceutical executives, and established neuroscientists, including Jeff Dean (the head of Google AI), Randy Schekman (Nobel laureate leading the ASAP Parkinson’s project), and Ed Boyden (one of the founders of optogenetics). We’re also supported by the Michael J Fox Foundation and, of course, the Sergey Brin Family Foundation.

I grew up in Germany, where schools are notoriously hard and teachers don’t tend to praise their students. Back then, it was hard for me to believe that I was smart enough to accomplish anything. During my undergraduate studies in Austria, I was able to get a fellowship that sponsored me to do full time research in the US. During my final year, I was living in a mattress inside a walk-in closet.

Later, when I graduated as the first ever Ph.D. in Stem Cell Biology from Stanford, in a record time of 2.5 years, I was all set for a promising career in academia, when my life turned upside down. After receiving the news of my mother’s illness, I decided instead to start a company to pave the way to curative treatments for complex diseases.

I had no money, my US visa was expiring, and I had no experience in business and technology. In the first year, I couldn’t persuade any engineer to come on board and didn’t have have a prototype to show investors. I was couch-surfing and sleeping in my car. With my roommate, we built our first product and got our first investment. That changed everything.

When I started, I was able to hire a small team, but shortly after I realized not only was our approach wrong, but my team was also wrong. We simply didn’t have what it took to create a map of all brain-aging disorders to identify and develop cures. Our initial approach was quickly leading us nowhere. Furthermore, my co-founder was not the right person for the role. He was manipulative, creating a toxic culture that was leaving many of us unhappy. At that point, I didn’t have a viable product, the team was wrong, and the money was quickly running out. In a last ditch attempt, I decided to re-start with a team with the right science and engineering background, we erased all code that had been produced and started from scratch. It was a scary moment that ultimately saved the company. Today the core team is still in place, our culture is thriving and we’ve built a machine learning platform that’s changing an entire industry.

There is no other platform out there thinking about information in this way. We are on the cusp of scientific innovation, creating new ways of analyzing information and solving problems.

Our opportunistic business model

Our technology allows us to pursue multiple avenues of monetization, which is crucial, as drug discovery timelines are long, and approval processes are time-consuming and expensive. Compared to companies whose machine learning approach is geared towards one particular data type or more traditional biotech companies that bet all their resources on one target or one molecule, our approach is significantly less risky for investors. 

On our path of becoming the first company to bring disease-modifying treatments for brain-aging diseases, such as Parkinson’s and Alzheimer’s disease, to market, we have options to generate cash flow in the short run and therefore mitigate our risk compared to many other biotechnology companies. The results provided by our platform can be generated for many diseases. Hence there are synergies in partnering with pharmaceutical companies as they can provide us with the capital while we generate unique proprietary insights into a disease area outside Parkinson’s. 

Identifying disease targets and mechanisms and reducing the risk of failure due to efficacy is a core interest for pharmaceutical companies. Even if we reduce the failure rate by as little as 10%, it will result in billions of dollars in savings for the pharmaceutical companies. Not surprisingly, large and mid-sized pharmaceutical companies have approached us to work with us.

We have received grants from the Sergey Brinn Foundation and the Michael J Fox Foundation. We have also raised over $5 million dollars to date and continue to raise funds as necessary in the future.