Joshua Sammet

Interpretability in deep neural networks for gene transcription prediction in Alzheimer's disease

I am a member of the AI team in the Computational & Molecular Neuroscience laboratory. My main interest is using interpretability methods for AI to better understand the neurobiology and genetics of psychiatric diseases.

My DPhil research, supervised by Prof Alejo Nevado-Holgado, Dr Sally Cowley, and prof Mark Fiers (KU Leuven), focuses on interpretability methods for Genomic foundation models (GFMs) to gain insight in cellular responses to Alzheimer's disease.
Using explainability methods, we are investigating how genetic mutations lead to changes in the predictions of these models and investigate how this can be applied to facilitate our understanding of the biological mechanisms behind it. I want to apply this insight to better understand differences in microglial responses to the pre-clinical stages of Alzheimer's disease.
This research is funded by the Berrow Foundation Lord Florey Scholarship.

I hold an B.Sc. in Computational Science & Engineering, and an M.Sc. in Biomedical Engineering from ETH Zurich. My previous research spans neurobiology of psychiatric diseases, LLMs in healthcare & medicine, and mathematical modelling of biological processes.