Simply Put
Every cell relies on complex genetic “instruction manuals,” called gene regulatory networks (GRNs), which can break down as we age or develop diseases. Currently, scientists lack effective computational tools to accurately compare how these networks differ between groups, such as younger versus older individuals. This project will build a powerful AI-driven framework to analyze advanced single-cell data and pinpoint exactly how these genetic instructions change across different conditions. To prove this tool works, the team will map the ageing process of the human thymus, a crucial organ that makes immune cells and test their findings in lab-grown “mini-organs.” Ultimately, this research will provide a highly accurate tool for disease modeling and uncover new targets to target the immune system, for instance to rejuvenate the aging immune system.
Detailed Description
The identity and function of a cell are fundamentally defined by unique signatures of gene expression, which are controlled by complex gene regulatory networks (GRNs). In this context, the onset of aging and disease phenotypes is frequently driven by the progressive deregulation of these networks, such as those leading to compromised chromatin accessibility and thymic involution. Due to the high complexity of these context-specific networks and their combinatorial interactions between transcription factors (TFs), cis-regulatory elements (CREs), and target genes, modeling GRN deregulation requires lots of detailed single-cell multiomic data.
In this project, a novel Generative AI (GenAI) model for multiome data will be developed to predict and compare differential GRN usage across biological conditions. To train this model, we will assemble a single-cell multiomic atlas of the human thymus spanning over 60 donors from fetal, pediatric, and adult stages, with paired transcriptomic and epigenomic data points per cell.
This dataset will be used:
- to train the GenAI to learn latent representations of T cell development and age-related immune deterioration
- to work out an optimal architecture for handling technical and biological covariates within the neural network
- to define the most relevant TF-CRE-gene triplets throughin silicomodeling of regulatory interactions
- to design a predictive framework for the downstream effects of age-induced multiomic changes
After training, the GenAI model will be deployed to predict the specific effects of aging on cell type-specific GRNs. Advanced generative techniques will be used to estimate causal relationships through counterfactual predictions by asking ‘what would happen to this cell if a key regulatory switch has never been turned off?’ This enables the computational prediction of how age-related epigenetic shifts directly alter GRN topology.
Initially, the AI analysis will rely on observational aging data to investigate the in silico effects and the relative contribution of each regulatory element to the aging process. Further, these predicted causal features will map exactly how specific transcription factors become deregulated over time, creating a unified model that captures nonlinear, synergistic effects of aging on the GRNs.
After the GenAI model and its aging predictions are established, its accuracy will be rigorously investigated using biological interventions. Specifically, an artificial thymic organoid system will be used to introduce targeted CRISPR and chemical perturbations. This will replicate the AI’s findings in vitro, validating the causal dynamics of GRN deregulation and confirming the model’s predictive power outside of a purely computational environment. In doing so, the project will yield new biological insight into immune aging, including which molecular switches, if restored, could help the aging immune system produce more of the T cells that protect us from infection and disease in later life.