CRISPR-EpiGEM: AI-Powered Discovery of Epigenetic Targets

Background

Regenerative medicine and targeted cellular therapies rely heavily on the ability to precisely guide stem cells into specialized cell types, such as bone, fat, or cartilage. A critical mechanism governing this cellular differentiation and lineage commitment is DNA methylation, an epigenetic modification that regulates gene expression without altering the underlying genetic sequence. With the emergence of CRISPR-based epigenetic editing tools, researchers now possess the capability to selectively add or remove methylation marks to control cell fate and treat diseases characterized by aberrant epigenetic states. Consequently, there is an urgent need for robust methods to accurately map and select the exact genomic loci that must be modified to drive desired phenotypic outcomes.

However, realizing the full potential of precision epigenome engineering is severely hindered by the immense, high-dimensional complexity of the human epigenome. Existing analytical approaches struggle to pinpoint the specific, functional cytosine-phosphate-guanine (CpG) sites whose methylation status causally drives cellular differentiation amidst hundreds of thousands of non-functional sites. Furthermore, standard computational frameworks often fail to capture complex, non-linear, and long-range dependencies across the genome without incurring prohibitive computational costs, while technical batch effects across different datasets frequently obscure true biological signals and impede effective target discovery.

Technology

CRISPR-EpiGEM is a deep learning framework designed to identify functional DNA methylation (DNAm) targets for CRISPR-based epigenetic editing. Built around a contrastive variational autoencoder (cVAE), the system processes high-dimensional Illumina array data to learn non-linear, cell-type-specific representations of DNAm profiles. Its encoder utilizes a Linformer-style CpG attention mechanism to map profiles into a low-dimensional latent space, while the decoder reconstructs the profiles for in silico perturbation simulations. By applying feature attribution analysis, the platform calculates latent representation gradients to pinpoint specific CpG sites for targeted silencing (via CRISPRoff) or activation (via dCas9-TET1). Additionally, the solution integrates an experimental workflow utilizing the RALA cell-penetrating peptide system to deliver epigenetic editors into mesenchymal stem cells with high transfection efficiency and minimal toxicity.

This technology is differentiated by its ability to overcome the high-dimensional complexity of the epigenome, solving a critical bottleneck in regenerative medicine by pinpointing the exact CpG sites that causally drive cell lineage commitment. Unlike standard transformers that suffer from quadratic computational costs, its Linformer-style attention mechanism efficiently captures long-range genomic dependencies. Furthermore, the integration of a multi-component loss function—specifically utilizing contrastive loss—uniquely structures the latent space to maximize the separation between distinct cell phenotypes while minimizing batch effects. This enables exceptional biological fidelity and generalization across independent datasets, providing a cost-effective, data-driven platform that uniquely bridges deep generative modeling with practical, non-toxic in vitro delivery.

Benefits

• Accurately identifies specific DNA methylation targets that drive cell phenotype and lineage commitment
• Processes high-dimensional epigenetic data with reduced computational cost using an efficient attention mechanism
• Minimizes batch effects and generalizes effectively across independent datasets
• Enables in silico perturbation simulations through generative modeling of methylation profiles
• Discovers both methylation and demethylation targets for precise gene silencing or activation
• Delivers epigenetic edits to primary cells with high transfection efficiency and low metabolic toxicity
• Provides a cost-effective, data-driven platform for optimizing stem cell differentiation and developing targeted epigenetic therapies

Applications

• Regenerative medicine stem cell engineering
• Epigenetic drug target discovery
• Targeted osteoarthritis therapy development
• Precision oncology epigenetic editing
• Automated cell differentiation optimization

Additional info

Tomas Gonzalez-Fernandez’s research focuses on engineering precision cell and gene therapies for regenerative medicine. The Tomas Gonzalez-Fernandez Lab (The TGF Lab) focuses on three interconnected areas: (1) applying synthetic biology tools such as CRISPR gene editing and epigenetic editing to engineer precision stem cell products with improved therapeutic capacity, (2) developing deep learning models for rational CRISPR target discovery and predictive cell reprogramming, and (3) designing biomaterial-based delivery systems for spatiotemporally and stimuli-responsive controlled gene editing in vivo and ex vivo.