PHOTOGRAPHY: MATHEW
PHOTOGRAPHY: MATHEW SCOTT I f you want to cure a disease, you first have to understand a disease. But how can you hope to know everything about it when the information is distributed across thousands of medical journals published over 50 years? It’s too much for any human to process. But artificial intelligence can now offer an answer. Medical research entrepreneur Katharina Volz is building a computer super-brain that can support human researchers by processing the vast amount of research data on a disease to suggest links and solutions that humans might never have spotted. Volz is CEO of San Francisco-based OccamzRazor, a company founded to help find a cure for Parkinson’s, a complex neurodegenerative disease that affects more than ten million people across the world. After graduating from Harvard Medical School and getting a PhD from Stanford, Volz was set to follow an academic career. But, in 2016, she received a call from someone close who had been diagnosed with Parkinson’s. “I was devastated,” she says. “I made a resolution that I was going to find a cure.” Eighty per cent of all the information in biomedicine is in medical publications and clinical documents. In order to start her search, Volz had to build a new technology that was able to understand these scientific papers to help formulate better hypotheses. “Parkinson’s is a poorly understood disease,” she explains. “Patients develop different symptoms – tremors, loss of sense of smell, balance problems. We need to bring together information from disparate areas like genomics, proteomics and metabolomics. It is too complex for humans to grasp “I MADE A RESOLUTION THAT I WAS GOING TO FIND A CURE FOR PARKINSON’S DISEASE” MEDICAL ARTIFICIAL INTELLIGENCE KATHARINA VOLZ CEO, OCCAMZRAZOR everything involved.” So Volz and her small team of machine-learning experts, biomedical scientists and computational biologists started by identifying the hundreds of key phrases to do with the disease, before working with the AI lab at Stanford University to develop a machine learning system that could ‘understand’ the papers. “We developed a natural languageprocessing algorithm that could identify various elements as well as their biological relationships,” explains Volz. Human Parkinsome, as the system is now known, is ten times faster than a human doing the same job and just as accurate. “We’ve now fed more than 20 million scientific papers through the system and, as soon as a new paper gets published, it’s immediately ingested into our system,” says Volz. Researchers can query the suggested hypotheses and feed the results back in, making the system smarter. Now that an AI brain has been developed to understand Parkinson’s, the team can apply the same technology to other diseases. “The platform is built in such a way that we can plug in data for cancer, for Alzheimer’s, for any complex biological disease,” says Volz. While these are early days, she is ambitious about changing the way we do medical research: “It’s ridiculous that email software is more intelligent than the software we use to find cures for diseases. But I feel hugely optimistic. My motto is: ‘Everything is possible’.” 49
