Concord-Carlisle Regional High School
CellProfiler, developed by the Broad Imaging Platform, is an open-source software that allows researchers to quantitatively measure phenotypes from thousands of cell images to study various biological questions. The current cell profiling method is roughly 80% accurate at determining the mechanism of action of unknown drug treatments. To improve this accuracy, Santiago was interested in testing three methods of profiling images of cells and analyzing their effects on downstream analysis. Using a known dataset from an experiment that tested drug treatments on breast epithelial cells, Santiago tested neural networks to perform segmentation analysis, as well as an inception-ResNet architecture to extract convolutional features. The results revealed that segmentation using neural networks can improve accuracy to ~85%, and up to ~90% when extracting convolutional features. These data suggest that deep learning can increase accuracy of cell profiling for predicting the mechanism of action of unknown drugs.
Santiago applied to BSSP because he wanted to go beyond the classroom for learning about the things he was interested in the most. He felt the Broad was a perfect opportunity for him to apply his existing knowledge to real-world problems and to ultimately learn so much more. "My time at the Broad, I believe, has been the best learning experience I’ve ever had, and the experience I am taking away will help immensely in my personal projects and in starting my career," said Santiago.