The human microbiome plays a critical role in health and disease, but extracting reliable biological insights from multi-omics datasets remains challenging due to high dimensionality, sparsity, noise, and compositionality. This project proposes the development and application of graph neural network (GNN) models to enhance the reproducibility of microbiome analyses. Using structural learning techniques in PyTorch Geometric, the project aims to build graph-based models that more reliably capture underlying biological patterns than traditional correlation-based methods. Beyond applications to the microbiome, the approaches developed through this project have potential for broad applicability to other domains involving complex biological networks, such as cancer genomics and immunology.

I am participating in SuperUROP to expand my foundation in experimental biology into computational research. My experiences with research in synthetic biology through MIT iGEM and cancer immunology at Dartmouth-Hitchcock have prepared me with strong biological intuition to guide hypotheses. I am excited to gain hands-on experience with machine learning in a biomedical context and to connect computation with biology.