Patricia Suriana: Application of deep learning to enzyme classification problems
Through the generous support of the Stanford GRIP program, I did a 3-month internship from June-August 2022 at the AG Noé group, Freie Universität (FU) Berlin under the supervision of Prof. Frank Noé. The AG Noé group is larger than my group at Stanford and they also consist of multiple subgroups, in which the projects between subgroups can greatly differ. During the internship, I was working closely with the protein design subgroup, which research is largely related to my research at Stanford. At Stanford, I work on the application of deep learning to molecular structural biology, in particular to structure prediction and drug design.
During the summer, I worked on application of deep learning to enzyme classification problems. Although related to my PhD research, the problem touches topics in enzyme/protein design which I have not worked on before. As we know, enzymes are important biological polymers that catalyze various chemical reactions. Enzymes are mostly proteins, excluding some catalytic RNA molecules or ribozymes. They are also highly efficient and extremely selective catalysts. There are multiple schemes developed to classify enzymes. Enzyme Commission (EC) number is one of them. It is a numerical classification scheme based on chemical reactions being catalyzed by the enzymes. EC number does not specify enzymes but enzyme-catalyzed reactions; it does not consider amino acid sequence (i.e., homology) or the protein structure. Hence, different enzymes might have the same EC number if they catalyze the same reaction.
Predicting EC number, which entails better understanding of enzyme activities/functions, has been an interest in the biology community. EC number prediction is a hierarchical (tree) multi-label classification problem as it consists of 4 digits ID, where each goes progressively to finer level, and that one enzyme can be annotated with different EC numbers (i.e. catalyze multiple reactions). Several deep-learning methods, either sequence-based or structure-based, have been explored. In this project, we explore a structure-based EC number prediction method. As have been shown in previous works, structural data contains more information compared to sequence. In addition, enzymatic functions and properties of a protein are also closely related to its structure. Preliminary results of our deep learning model trained on structural enzyme data look promising. Beyond the internship, I plan to continue work on this project and continue collaborating with the team at AG Noé.
Overall, I had an amazing and rewarding experience working in the AG Noé group. In addition to learning more about enzyme and computational protein design, I had the opportunity to interact with other amazing people working in different subgroups and learned more about their research. The AG Noé group also has regular Friday beer night, which I found pretty nice to get to know more people and learn more about the work/social life of PhD students in Germany.
Outside the internship, I had a great time in Berlin. Berlin is an amazing city for a classical music and opera lover like me; they have cheap opera tickets available for young adults. Berlin also has three opera houses: Deutsche Oper, Staatsoper Unter den Linden, and Komische Oper. I probably went to almost all the opera concerts played in Deutsche Oper and Staatsoper Unter den Linden while I was in Berlin. In addition, I also had the chance to go to Verona, Italy to watch the Arena di Verona Opera Festival. Although I rarely had the opportunity to practice my German beyond ordering bread at the bakery (since AG Noé and Berlin itself are pretty international), Berlin is also an amazing place for foodies. They have a great selection of bakeries and restaurants from around the world.
Overall, I thoroughly enjoy my time in Germany. Not only was I able to learn more skills and build the connections relevant for my PhD studies and future goals, I was also able to immerse myself in German culture, in particular the German classical music culture. It was such an enjoyable and rewarding experience and I highly recommend the GRIP program to any future students. Lastly, I would like to thank The Europe Center and the Stanford Club of Germany for the generous support to make this internship possible.