Develop machine learning models to study cell-type-specific aging using single-cell methylation data in the Uzun Lab
PI: Yasin Uzun (Penn State College of Medicine)
The remainder of postdoc salary will be supported through PI’s postdoctoral fellowship or Uzun Lab resources.
Project Description
Aging is a complex biological process controlled by both the genetic and epigenetic factors. Epigenetic marks such as DNA methylation have proven to be reliable biomarkers of biological aging, leading to the development of “epigenetic clocks.” However, most current models rely on bulk tissue data, which masks cell-type-specific variation and fails to capture early epigenetic changes in aging-related diseases such as neurodegeneration, immune dysfunction and cancer.
Our goal is to develop a deep learning-based framework to predict cell-type-specific epigenetic age using single-cell methylation data. Modeling aging at this resolution will allow us to detect lineage-specific aging trajectories, uncover early signs of dysfunction, and ultimately inform precision diagnostics and targeted interventions.
Core Objectives
1. Model Cell-Type-Specific Epigenetic Age: Develop and train deep neural networks (DNNs) on CpG-level methylation profiles to predict biological age at the single-cell level, across diverse cell types.
2. Detecting Lineage-Specific Aging in Disease Contexts: Compare predicted age across health and disease conditions (e.g., normal vs. Alzheimer’s or cancer tissues) to identify lineages that are exhibiting accelerated or retarded aging and enable detection of potential therapeutic targets.
3. Characterize Aging Heterogeneity Across Cell Type: Quantify deviations from chronological age to pinpoint which lineages age faster or slower, providing insights into cell-intrinsic aging programs.
4. Tool Development: Open-Source Software for the Community, build a modular and extensible python package for predicting and visualizing aging at the cell-type level, with downstream applications in both basic research and clinical translation.
Key Data Resources
- scEpiAge: Mouse peripheral blood single-cell methylation dataset
- GEO & ENCODE: Aging-related single-cell methylation data in mouse and human
- Single Cell Portal: Supplementary datasets on human aging
- ArrayExpress: Bulk and pseudo-bulk methylation benchmarks for comparison
Significance and Novelty
While existing epigenetic clocks estimate biological age, they do not necessarily generalize across cell types or utilize single-cell resolution. Our approach is novel in its solution of lineagespecific prediction and application of deep learning for the representation of complex CpG-age associations. This would enable detection of subtle, disease-related epigenetic shifts. This celltype-resolved framework will uncover aging mechanisms obscured in bulk data, and could directly inform strategies to modulate aging or delay disease onset in high-risk cell populations.
Project Needs & Junior Researcher Contribution
We are seeking 1 ICDS Junior Researcher with expertise in: Deep learning (PyTorch or TensorFlow), Single-cell omics, Epigenetic/aging analysis
Junior Researcher Tasks:
Design and optimize DNN/CNN architectures for CpG-age prediction, run experiments across multiple datasets (mouse and human), apply feature attribution (e.g., SHAP) to interpret aging signals, visualize cell-type-specific and disease-specific aging patterns, develop and document a reusable, scalable analysis pipeline
Proposed Effort and Funding Plan
12 months at ~25% RA effort. The remainder of RA salary will be supported through PI’s postdoctoral fellowship or Uzun Lab resources
Lab resources
(GPUs, servers) already available via ICDS and College of Medicine
Medium to Long-Term Goals
Submit full grant proposal (e.g., NIA/NIH or NSF AI-Driven Biosciences) based on validated findings, apply the model to disease-specific single-cell methylomes (e.g., cancer, neurodegeneration), extend work to multi-omics integration for systems-level understanding of aging
ICDS Mission Alignment
This project directly aligns with ICDS’s mission by applying AI and data science tools to biological data to answer critical questions about aging, disease, and health.
ICDS Engagement
Dr. Karamveer (postdoc, Uzun Lab) is actively engaged with the ICDS community through utilization of ICDS’s research infrastructure. The proposed project aims to further deepen this engagement through mentorship of ICDS Junior Researchers and shared tool development.