Brookline High School
Knowing the mechanism of action of drugs is essential for understanding how drugs work on a cellular level, which can be useful rationally designing new, targeted therapeutics. Yuen Ler spent his summer in BSSP programming a machine learning algorithm called a variational autoencoder in order to predict the mechanism of action of various drugs. A variational autoencoder uses several neural networks to take very large datasets, compress them down into a very low-dimensional representation, and then decompress them back into a large dataset. By performing arithmetic on this low-dimensional “latent space”, it is possible to combine inputs from several different experiments to produce an output that gives the predicted result of an experiment that hasn’t been performed yet. Yuen Ler successfully trained a variational autoencoder on cell morphology data obtained from fluorescence labeling experiments, and used it to predict how various drugs might affect the morphology of cells in order to determine their mechanisms of action.
“I have been interested in science ever since I was a kid, but my interest really kicked off in high school when I learned a lot more in school and on my own,” said Yuen Ler. “This computational biology summer experience has strengthened my decision to major in a STEM field, especially computer science.”