Chris Sander at Bio-IT World
By Bio-IT World Staff
April 23, 2015 | Chris Sander, chair of the Computational Biology Program at Memorial Sloan-Kettering Cancer Center, delivered Wednesday’s keynote address this week at the Bio-IT World Conference & Expo in Boston.
At Memorial Sloan-Kettering, Sander works on models of cancer to predict drug responses, with a particular emphasis on combinatorial therapies. That focus is a response to modern oncology’s central dilemma: new targeted therapies can have incredible short-term efficacy in fighting cancers with specific molecular profiles, but the varied and evolving nature of the disease means that tumors usually recover from the initial blow.
“The good news is that targeted therapies have done amazingly well since about the year 2000, with Gleevec being the poster child, of actually addressing the particular genetic alteration [that causes a patient’s cancer],” said Sander in his address. The bad news is that these cancers often develop resistance within a few months of treatment, thanks to secondary mutations that are unaffected by therapy. “The idea is to use more than one drug, more than two, and block the exits before the cancer escapes.”
To create computational models that suggest new drug combinations for specific subtypes of cancer, Sander’s department relies on “perturbation biology,” an analogy to the “perturbation physics” practiced with particle colliders. The method combines repeated disruptions to a system — impacts between particles, say, or compounds introduced to cell cultures — with exhaustive measurements to provide the raw material for new models. It’s an appropriate metaphor for Sander, who began his career in science as a theoretical physicist and moved to biology when he saw that developing fields like protein structure would demand orders of magnitude more computational power.
In his work on combinatorial therapies today, Sander seeks to build models of what he referred to in his address as “the main essence” of a cancer’s biology, the integration of a few key protein pathways that play a disproportionate role in the tumor’s ability to survive and grow. This can be a fine balancing act, as the scale of modern DNA and protein experiments can produce hundreds or thousands of variables to play with. “You want to keep things simple,” said Sander. “If you have too many parameters, if you have 10,000 numbers and only 2,000 experiments, you might as well forget it. You don’t want to overfit.”
This is the art of computational biology, Sander said, adding, “This is not just about compute… You have to do thinking along the way.” Successfully navigating these models, and testing and refining them over time, can produce predictions of two or more molecular targets to attack at once, to prevent resistance from ever developing. Sander’s group has predicted multiple drug combinations whose efficacy was later validated in preclinical studies at Memorial Sloan-Kettering, and is hoping to move some of these therapies into clinical trials.
While Sander spent the bulk of his keynote address on combinatorial therapy, he also touched on two other current projects he’s proud of. The first is cBioPortal, developed by Sander’s department at Memorial Sloan-Kettering and now expanding into a multi-organization open source project. This portal is a visualization and analysis tool for cancer datasets from large projects like The Cancer Genome Atlas, designed to clarify the key activities in cancer cases.
Sander showed the audience at Bio-IT World a particularly clever visualization from cBioPortal, displaying three tumor genomes in parallel, each presented as a map with areas of mutation highlighted. All three had been taken from the same patient at different time points, making clear how drastically cancers can change over time.
Finally, Sander briefly discussed his collaborations with Debora Marks at Harvard Medical School, which have brought him back to the beginning of his career in biology when he was focused on predicting protein structure. In the early 1980s, Sander helped to create DSSP, still a standard tool for predicting secondary structures of amino acids. Today, Marks and Sander together have been able to go much further, creating algorithms that predict proteins’ three-dimensional structure with high accuracy using only the amino acid sequence. (Publications connected to this work can be found here, here, and here.)
While these algorithms are currently limited to only a small subset of proteins, Sander is optimistic that their scope will expand rapidly as new structural data gives molecular biologists a better grasp of the physical constraints on protein conformation. “This has been an unsolved problem now for thirty years,” he noted. “It’s one of the most exciting developments that I’ve been involved in.”
More from the Bio-IT World Conference: