Research area: precision medicine for cancer

Exascale computing for targeting cancer

Shown left is a histology slide from a patient-derived xenograft (PDX) with a resolution of approximately 40,000 x 40,000 pixels. Shown right is a random patch of size 256 x 256 pixels from the PDX slide. Processing a single PDX slide using a convolutional neural network requires splitting the slide into small overlapping patches (approximately 90,000 patches are extracted for the shown example). A representative dataset of 2,000 PDX slides provides a training set of 180 million patch images. Credit: Alexander Partin, Cancer Distributed Learning Environment (CANDLE) project

About 40 percent of men and women in the United States will be diagnosed with cancer during their lifetimes, and treatment and prevention of the disease is a national priority. To develop new treatment strategies, doctors and scientists need to understand the key molecular mechanisms that lead to the mutations that cause cancer; compute highly accurate models for predicting drug response and effectiveness; and statistically analyze the progression of the disease in patients to identify patterns. And the faster these steps can happen, the better.

Scientists already are preparing computer codes and tools for the power of exascale systems. At exascale, researchers plan to simulate key protein interactions in RAS and RAF gene mutations, which are present in nearly one-third of cancers. Fifty times more powerful than today’s supercomputers, exascale systems can help develop predictive models to optimize precancer screenings and treatment options through precision medicine that match the best treatment to an individual’s type of cancer diagnosis. Researchers can also leverage new data analytics and deep learning techniques to extract information from millions of cancer patient records to determine optimal treatment strategies.