Alzheimer’s Disease Analysis
Project Overview
This research project focuses on the early prediction and molecular characterization of Alzheimer’s Disease (AD) using single-cell genomics data. By leveraging the ROSMAP (Religious Orders Study and Memory and Aging Project) dataset, we aim to identify robust disease markers that traditional analysis might miss.
Methodology
The core of this work involves moving beyond standard statistical tests to apply advanced Deep Learning architectures to biological sequences.
- Foundational Models: We are fine-tuning Large Language Models (LLMs) adapted for single-cell data, such as scGPT, Geneformer and others. These models treat gene expression profiles as “language,” allowing us to transfer learning from massive datasets to the specific context of Alzheimer’s.
- Machine Learning Benchmarks: We compare these deep learning approaches against traditional ML baselines (such as Random Forests and SVMs) to quantify the performance gain offered by transformer-based architectures.
- Classification Strategies: A key part of this study involves experimenting with different classification heads on top of the pre-trained models to optimize for accuracy and interpretability.
Tools & Technologies
- Data: ROSMAP (Single-cell RNA-seq)
- Models: scGPT, Geneformer, few other TBD
- Frameworks: PyTorch, HuggingFace
- Techniques: Fine-tuning, Transfer Learning, Explainable AI (XAI)