Machine Learning

Machine learning methods have proven to be amazingly powerful for a wide variety of problems.  Successes have catalyzed much excitement for applications to new areas and for continued advances. Traditional approaches are typically most valuable when very large amounts of data are available.  However, many problems in biology and medicine do not have vast amounts of data of the type required to classically train a ML model.  Much of biological and medical knowledge has been acquired in a piecemeal manner, and although important and powerful, is not classically implementable into ML.

We are very interested in developing informed machine learning methods and models that combine prior, traditional, biological knowledge into the ML workflows.

As a combined experimental and computational lab, we extensive expertise with the various forms and uses of traditional biological knowledge.  The lab also has extensive expertise with the development of mechanism-based mathematical models, and with the complex, high-dimensional, relationships between biological “features” and biological behaviors.

We are working to combine our various strengths to develop machine learning models where traditional biological knowledge is more fully infused throughout machine learning development with the goal of developing models that are more powerful and more predictive while also requiring less training data.

Several of our current projects in this area involve RAS, the RAS pathway, and cancer as there is extensive prior knowledge and mechanistic information, and because we have a long history of mathematical modeling with this information.