Scientists at Columbia University reported that they had developed a new computational model predictive of breast cancer survival. The Columbia Engineering researchers, led by Dimitris Anastassiou, Ph.D., Charles Batchelor Professor in electrical engineering and member of the Columbia Initiative in Systems Biology, won the Sage Bionetworks / DREAM Breast Cancer Prognosis Challenge for this work, published in the April 17 issue of Science Translational Medicine.
The authors trained their model using genomic and clinical data from more than a thousand women diagnosed with breast cancer, then tested it using a new dataset from 184 women also diagnosed with breast cancer. They found that it could best predict breast cancer survival compared to the other models in the Challenge.
In earlier work, Anastassiou and his team identified what he calls “attractor metagenes,” gene signatures that are present in nearly identical form in many cancer types. Working with doctoral students Wei-Yi Cheng and Tai-Hsien Ou Yang, he tested the signatures in the Sage Bionetworks/DREAM Breast Cancer Prognosis Challenge, a cloud-sourced effort for accurate breast cancer prognosis using molecular and clinical data. The team developed a prognostic model that showed that these signatures of cancer, when properly combined, were strong predictors for breast cancer survival.
The researchers believe their gene signatures and computational model could together improve cancer diagnostic signatures and spur the development of new treatments that block the molecular mechanisms responsible for making cancers aggressive or invasive.
Dr. Anastassiou commented on the breast cancer work, “These signatures manifest themselves in specific genes that are turned on together in the tissues of some patients in many different cancer types. And if these general cancer signatures are useful in breast cancer, as we proved in this Challenge, then why not in other types of cancer as well? I think that the most significant—and exciting—implication of our work is the hope that these signatures can be used for improved diagnostic, prognostic, and eventually therapeutic products applicable to multiple cancers.”
The goal of the Sage Bionetworks/DREAM Breast Cancer Prognosis Challenge is to assess the accuracy of computational models to predict breast cancer survival based on gene expression data, copy number data, and standard clinical information of 2,000 breast cancer patients (the MATABRIC dataset). The challenge was structured as a scientific collaborative competition in which participants were encouraged to use a common software platform (Synapse) for model submission and assessment.
Commented Sage Bionetworks founder Stephen Friend, Ph.D., and one of the organizers of the Breast Cancer Challenge (BCC), “Ten years ago, members of our research group used gene expression profiling to build one of the first breast cancer predictors. Mammaprint and Oncotype Dx were developed off of that but further improvement seems to have stalled. We wondered if running a Challenge like BCC would motivate lots of different groups to tackle this problem working collaboratively, and if that might be more fruitful than the current ‘go it alone’ single researcher approach.”
Anastassiou and his team noted that the BCC provided a vibrant research environment where numerous participants were openly submitting their models and had access to others’ models as they were developed. The teams were also encouraged to incorporate the other models into their own. Three hundred fifty-four participants from more than 35 countries registered for the Challenge and submitted more than 1,700 models.