Using artificial intelligence to advance personalized medicine
Opened less than a year ago, the Swiss Data Science Center – a joint initiative between EPFL and ETH Zurich – has already launched eight research projects in fields ranging from personalized medicine and environmental protection to open science. Each project brings together experts from several disciplines to join forces in tackling some of society’s biggest challenges.
One of the eight research projects, involving personalized medicine, stands to revolutionize treatment options for cancer patients. It is being carried out by EPFL’s Signal Processing Laboratory (LTS4) in association with Lausanne University Hospital (CHUV). The project scientists are using advanced machine-learning algorithms to analyze images of tumor cells and identify the distinguishing features of each type of tumor. That will enable doctors to better categorize patients’ tumors and select the most effective treatment for each one, broadening the range of personalized therapies available. This type of information, which has not yet been widely exploited, will be obtained by automatically analyzing vast amounts of data. “Personalized medicine is a rapidly growing field, thanks largely to high-speed sequencing,” says Olivier Michielin, chief physician of the analytic personalized oncology division at the CHUV. “Today, oncologists choose cancer treatments based in part on the shape of the tumor as seen under a microscope.” And while shape can be useful for initially classifying a tumor, the information generated from sophisticated, interpretable algorithms – combined with human judgment – can allow for a much more granular diagnosis.
Developing personalized treatments
Drawing on their expertise in machine learning, the EPFL scientists will develop automatically-run algorithms to process the data. These algorithms will train computer models to spot key tumor features based on a large data set. The goal is to develop methods that doctors can interpret and use to evaluate the results obtained from a computer. “If the model classifies a tumor in a given category, we need to know why – what aspect of the tumor’s architecture led to that decision,” says Michielin. “Machine learning, and especially the architecture currently used in deep learning, can give good predictions in many applications, but it’s not always easy to understand their results in detail,” says Pascal Frossard, head of the Signal Processing Laboratory. But when it comes to patient treatments, doctors must be able to thoroughly comprehend why a given conclusion was reached before they make any major decisions.
For this research, the project team will focus on melanoma, the most aggressive kind of skin cancer. They aim to be able to predict how the tumor will respond to immunotherapy, which is a treatment method that involves stimulating a patient’s own immune system to fight the cancer cells. “This type of treatment can be highly effective – but not on everyone. We want to understand why, and what tumor characteristics are typically associated with a positive outcome. That will help doctors select the best treatment for a particular patient,” says Michielin.
Drawing on several disciplines
Another Swiss Data Science Center project being carried out in association with EPFL involves developing an electronic platform to house data collected as part of the Antarctic Circumnavigation Expedition (ACE) project and to make that data available to the general public. During this expedition, some 150 scientists studied phenomena in the Antarctic to better understand how climate change is affecting the Southern Ocean and what that means for our entire planet. A third project aims to develop a platform for open science competitions where scientists from around the world can come together to tackle challenging problems like how to diagnose diseases in plants, predict a person’s size or teach a virtual skeleton to walk and run.
A national initiative
The Swiss Data Science Center is a national initiative designed to foster innovation in data and computer science and to provide the necessary infrastructure for promoting cross-disciplinary research and open science. It’s being implemented jointly by EPFL and ETH Zurich, with offices in Lausanne and Zurich. The initiative will enable Switzerland to develop expertise in data science and become a center of excellence with global standing.
iLearn: Interpretable Learning Methods for Immunotherapy is a research project being carried out jointly by EPFL’s Signal Processing Laboratory (LTS4) and the Oncology Division of the Lausanne University Hospital (CHUV).