Research Intern with EpiAgent Downstreams

Foundation models for single-cell epigenomics, such as EpiAgent, have recently demonstrated the ability to capture chromatin accessibility landscapes and perform in-silico perturbation tasks, including regulatory element knockouts in cancer. However, extending such frameworks to neurodegenerative diseases requires modeling brain-specific regulatory contexts and cellular heterogeneity. In this study, we expand the EpiAgent in-silico treatment paradigm from cancer to Alzheimer’s disease (AD) by incorporating cell-type specific low-rank adapters (LoRA). These adapters enable fine-grained modulation of the pretrained foundation model, allowing differential simulation of regulatory responses across major brain cell types, including neurons, astrocytes, microglia, and oligodendrocytes. Using single-cell ATAC-seq datasets from AD and control brain tissues, we demonstrate that cell-type specific adapters enhance sensitivity in detecting disease-relevant regulatory programs and improve the fidelity of in-silico treatment simulations. Our results suggest that integrating modular adapters into foundation models provides a scalable strategy to simulate therapeutic effects in a cell-type aware manner, opening new avenues for computational drug discovery and mechanistic insight into complex brain disorders.