BayesMendel Lab
The BayesMendel Lab develops sophisticated statistical models and software that identify individuals with elevated cancer risk due to inherited genetic factors, providing them with accurate risk assessments, and better access to detection and prevention strategies.
450 Brookline Ave
Boston, MA 02215
What we do
The BayesMendel Lab specializes in the development of predictive tools—including statistical models and software—related to cancer susceptibility genes, which are genes that can increase a person’s risk of developing cancer.
Based at the Dana-Farber Cancer Institute (DFCI) and the Harvard T.H. Chan School of Public Health and led by Drs. Danielle Braun and Giovanni Parmigiani, the BayesMendel Lab combines two powerful approaches:
- Statistical and machine learning methods that go back to Thomas Bayes and which help analyze probability and uncertainty;
- Genetic inheritance models established by Gregor Mendel, the father of modern genetics.
Our prediction algorithms have been used clinically for about 25 years and contributed to better cancer risk assessment and prevention.
Current Methodological Research
Our methodological research considers statistical and machine learning challenges that are critical for better developing and evaluating prediction models.
Examples of methodological projects include:
- Accounting for misreporting of family history in Mendelian risk prediction models.
- Using a frailty model to account for unobserved genetic and environmental effects in Mendelian risk prediction models.
- Accounting for missing data in Mendelian risk prediction models.
- Combining predictions from multiple risk models.
- Developing statistical methods to conduct meta-analysis of penetrance estimates.
- Developing statistical methods to conduct meta-analysis of prevalence estimates.
- Developing Mendelian risk models that consider multiple genetic syndromes together.
- Developing computational approaches to extend Mendelian models to large number of genes and cancers.
- Updating and extending existing Mendelian models, including; BRCApro, MMRPRO, PancPro, MelaPro, Fam3PRO, and LFSpro.
- Developing machine learning tools that can learn mendelian inheritance patterns.
What is inherited susceptibility to cancer
Cancer is caused by mutations in our genes. While many of these changes occur during one’s lifetime, some—called germline mutations—are inherited. Genetic research has identified many genes where inherited mutations carry a significantly increased risk of one or more types of cancer. Important examples are the BRCA1 and BRCA2 genes, responsible for the familial breast/ovarian cancer syndrome, and certain mismatch repair genes responsible for familial colorectal cancer.
Most mutations of cancer genes do not determine the fate of an individual. Genetic effects are therefore characterized probabilistically by penetrance functions, that is probability distributions of developing cancer by age, given a specific genetic variant, and prevalence functions, that is frequencies of mutation by population strata.
Identifying individuals at high risk of cancer because of inherited genetic susceptibility, and providing them with reliable assessments of cancer risk, is complex and increasingly important. Probabilistic prediction algorithms that exploit domain knowledge of Mendelian inheritance and other biological characteristics of susceptibility genes have successfully contributed to improved screening, prevention, and genetic testing.