Application of Sparse Autoencoders to Enhance Mechanistic Interpretability of Large Language Models in Medicine (Preprint) | Synapse