Regina Barzilay
Massachusetts Institute of Technology
- Research field
- Machine learning for medicine and chemistry, natural language processing
Regina Barzilay builds machine learning models for medicine and chemistry, among them models that read mammograms to estimate cancer risk and models that predict the properties of molecules from their structure. In a 2020 paper in Cell with James Collins and their colleagues, a neural network trained on 2,335 molecules that the laboratory had tested against the bacterium Escherichia coli was used to rank thousands of molecules under study for other diseases, and the top candidate, named halicin, turned out to kill a wide range of resistant bacteria by a new mechanism. The same model then searched more than 107 million molecules in four days and pointed to 23 candidates, of which 8 were active. The unit presents this work in its section on artificial intelligence as a tool of science, as a classifier of the kind Unit 5 taught that directed a laboratory to the corner of chemical space worth testing.
In the masterclass
Section 5, on artificial intelligence as a tool of science, presents the halicin discovery and works through its numbers.
The research is presented in Unit 8: The Big Questions of Artificial Intelligence, in a research spotlight box that explains the result at the level of the unit and links it to what you have just learned.