Aptamers are short single-stranded nucleotide sequences that fold into specific three-dimensional structures. With high specificity, they can bind to many disease-causing target proteins, making them a promising new therapeutic modality. Although aptamers are relatively short molecules comprised of approximately 30 nucleotides, there are trillions of possible sequences for an aptamer of this length, and we can only evaluate a very small fraction of these designs experimentally. This motivates the development of computational methods to design aptamers for new targets, sampling the most promising designs from this large space. In this work, we propose a structure-based machine learning method to design aptamers for new protein targets. Given that the volume of aptamer-protein complex data is very limited, we investigate how related data from small molecules, antibodies, and peptides can be harnessed for this application. We aim to cross-modally condition structure generation methods like diffusion and flow-based models to produce realistic RNA aptamer structures that can bind to a protein of interest.

Through SuperUROP, I hope to gain experience with methods development for computational biology. My past research projects have given me a strong grasp of machine learning fundamentals, especially in the context of molecules and proteins. Building on that, I’m excited about building methods for the challenging setting of nucleic acid structure prediction. Learning from my mentors, I hope to balance thorough experimentation with creativity.