Revere High School
Daniel King and Cotton Seed
Stanley Center for Psychiatric Disease
It seems obvious that the brain is the home of psychiatric disease; however, science has still not found a clinical-grade, biological marker for psychiatric disease. Jonathan, with his partner Ilora, hypothesized that biological markers are hidden among the immense complexity of brain images. He suspected machine learning techniques could distinguish the portions of brain images due to heritable traits (i.e. genetics) from those due to non-heritable traits (i.e. environmental factors). Specifically, Jonathan postulated that MRI images could be used to train a special type of neural network known as an autoencoder, which, when trained properly, would reduce all of the extremely complex phenotypes present in an MRI image down to a handful of patterns that make up these complex phenotypes, known as latent phenotypes. Jonathan’s project involved proving that, in principle, this idea could work; to do so, he trained an autoencoder using a dataset of faces that had been constructed using a deep convolutional neural network in order to see if heritability information about these faces could be captured by just a few latent phenotypes. His preliminary results were extremely promising, meaning that seeing someone’s genetic information from an MRI image may not be science fiction for much longer! When talking about why he applied to BSSP, Jonathan said “I was interested in the cutting edge biomedical research at the Broad Institute, and felt that this six-week program would capture what it means to be a scientist. The Broad’s global thinking attracted me to the community.” Jonathan's favorite part of being a Broadie was learning from a group of people who work in meaningful research with intentions of improving medicine for the world. "Working with my mentors has been a great pleasure and I have understood a lot from their expertise," said Jonathan.