Machine Learning Platform Identifies Unknown Genetic Drivers of Cancer to Personalize Treatment Strategies

Xinghua Lu, left, and Greg Cooper, right, of the University of Pittsburgh School of Medicine Department of Biomedical Informatics.

Xinghua Lu, left, and Greg Cooper, right, of the University of Pittsburgh School of Medicine Department of Biomedical Informatics.

Despite a decline in overall U.S. cancer mortality rates, cancer is still among the leading causes of death worldwide. Recent advances in immunotherapy have resulted in extremely effective cancer drugs that help the immune system do what it is designed to do: recognize and kill foreign invaders, including cancer cells. However, immunotherapies only work for about 30 percent of cancer patients, and it is unclear why some patients respond while others do not.1

Professors Greg Cooper and Xinghua Lu of the University of Pittsburgh Department of Biomedical Informatics are innovating new methods to improve cancer treatment strategies, starting with melanoma. What’s their strategy? Machine learning (artificial intelligence) tools to enable precision oncology.

Cooper and Lu have pioneered a machine learning platform, called Tumor-specific Driver Identification (TDI) that helps predict the effectiveness of cancer immunotherapies by identifying key genetic mutations that help cancer cells evade the immune system.
“For the first use case, we are investigating how melanoma tumors that fail to respond to treatment exploit the immune system. We think TDI will provide meaningful data to search for alternative therapeutic strategies and combination therapy opportunities.”

Xinghua Lu, PhD, University of Pittsburgh School of Medicine Department of Biomedical Informatics.

Improving immunotherapy by understanding what drives melanoma

More people are diagnosed annually with skin cancer than all other cancers combined,2 and one in five Americans will develop skin cancer by the age of 70.3 Melanoma, the most serious type of skin cancer, is estimated to cause more than 9,300 deaths in the U.S. in 2018 alone.4 After approximately four decades without new treatments, recent immunotherapy breakthroughs have led to eleven new melanoma drugs being approved by the FDA since 2011.4

Use of immune checkpoint blockades, such as CTLA-4 inhibitors (ipilimumab) and anti-PD-1 drugs (nivolumab and pembrolizumab), are widely used as an adjuvant therapy to treat melanoma in addition to surgical approaches.

Patient outcomes have shown that when immunotherapy works, it works really well and has even resulted in long-term remission for some patients with metastatic cancers. Yet, it is uncertain why almost two thirds of patients respond poorly to these drugs.

With TDI, Cooper and Lu hope to not only to identify the patients who will respond (or fail to respond) to a given treatment, saving time and money, but also to identify alternative therapeutic strategies that convert non-responders into responders. These alternative treatments can turn the immune system’s attack switch back on, allowing the body to fight cancer effectively.

One-size-fits-one vs one-size-fits-some

The majority of currently available methods aim to treat cancer by identifying genomic alterations found in a collection of tumors responsible for tumor growth.

It is common for a tumor cell to host hundreds of gene alterations; only a small minority of those alterations, known as “drivers”, help the cancer progress. Other mutations are often unrelated and known as “passengers.” If a certain mutation occurs frequently in certain type of tumor, then the mutated gene is assumed to be a tumor driver.

Types of cancer-causing gene alterations are well-known, such as chromosome structure variations, non-coding mutations, and epigenetic modifications;5 yet current approaches do not identify tumor-drivers based upon a combination of these alterations, nor do they estimate the functional consequences of a single gene mutation.

By studying tumors at the individual patient level using genomic sequencing, TDI can elucidate tumor-specific, mechanistic changes, related to molecular recognition and expression patterns. This robust approach will help oncologists to deliver the most effective, personalized therapies for each patient, perhaps identifying therapies that would not have been previously considered or by predicting combination therapies likely to improve outcomes.

Cooper further explains the benefits of being able to elucidate the cancer-driving mechanisms of an individual tumor in stating that:
“Tumors are highly heterogeneous in terms of their genomic drivers and their cell-signaling-pathway mechanisms; the better we can infer the specific mechanisms of a given tumor, the better we will be able to tailor precision therapies to treat that tumor.”

Greg Cooper, PhD, University of Pittsburgh School of Medicine Department of Biomedical Informatics.

If TDI can explain why patients do or do not respond to immunotherapy, the determining factors could be corrected in patients to extend the benefits of immunotherapy and the lasting responses it can provide to more people.

Collaborate to commercialize

Lu and Cooper have embraced collaboration to advance their translational research by teaming up with Pitt’s Center for Commercial Applications of Healthcare Data (CCA) and sciVelo (part of the Innovation Institute). The CCA, through the Pittsburgh Health Data Alliance, funded the technical development and validation of the TDI algorithm and initial commercial software development that was completed in December 2016.

The next phase of the project is being supported by significant follow-on funding from a joint Pitt/UPMC translational research program, the UPMC Immune Transplant and Therapy Center. This funding provides the opportunity to perform prospective clinical validation of TDI’s ability to predict melanoma patients’ response to immunotherapy.

Center for Commercial Applications of Healthcare Data

The Center for Commercial Applications of Healthcare Data, in collaboration with Pittsburgh Health Data Alliance, accelerates the translation of health data-focused technologies into products and services that save lives, improve health and reduce healthcare spending.  The Center offers collaborative support between Pitt’s renowned researchers and industry partners to transform breakthrough University innovations into critically needed healthcare solutions.  The Center for Commercial Applications of Healthcare Data is housed in the Department of Biomedical Informatics - a leader in improving research and clinical care at the intersection of medicine, biotechnology, and information technology.  Pitt sciVelo and the University of Pittsburgh Innovation Institute manage the Center for Commercial Applications of Healthcare Data and provide commercial translation support to accelerate the translational success of CCA-funded research projects.

Innovation Institute

The Innovation Institute was established in 2013 and is the University’s hub for innovation and entrepreneurship. The Innovation Institute provides a comprehensive suite of services for Pitt Innovators, from protecting intellectual property to the commercialization of new discoveries through licensing and/or new enterprise development. The Institute also provides a wealth of educational programming, mentoring and networking for Pitt faculty, students and partners.


sciVelo is part of the University of Pittsburgh Innovation Institute, accelerating life and health sciences translational research and commercialization. sciVelo advances Pitt’s translational research ecosystem through a transdisciplinary team of over 20 practicing scientists, clinicians and translational research experts. sciVelo’s mission is to Engage in Research of Impact by identifying, cultivating and advancing Pitt’s most promising life and health sciences translational research to market-ready solutions.

Immune Transplant and Therapy Center

Immune Transplant and Therapy Center is a bold and ambitious effort. A partnership between UPMC and the University of Pittsburgh, the center is revolutionizing how the world thinks about - and treats - many diseases.  This newly established venture focuses on clinical areas in organ transplant, cancer and biology of aging. With the creation of the Immune Transplant and Therapy Center, UPMC and Pitt bring their brightest minds in immunotherapy research to:
  • Explore the potential of immune transplant and therapy.
  • Design novel treatment approaches that harness or control the body’s natural defenses to fight harmful diseases and infections.
  • Share ideas and drive further development in treatments for cancer and other diseases.
With extended research and expertise and access to a robust series of clinical trials, the UPMC Immunotherapy and Transplant Center team is dedicated to bringing life-saving solutions to our patients as quickly as possible.

Author/Photographer: Alyssa B. Lypson, MS

In February 2018, a technical explanation of the TDI algorithm was published in bioRxiv - "Tumor-specific Driver Inference (TDI): A Bayesian Method for Identifying Causative Genome Alterations within Individual Tumors." To learn more, visit here.