Accelerates targeted therapeutics via in silico variant prioritization across 100+ cell types.

Background

Eukaryotic gene expression is intricately regulated by chromatin structure and epigenetic markers, which are governed by DNA sequence elements like promoters and enhancers. Genetic variations, such as single nucleotide polymorphisms, can significantly alter these regulatory landscapes and disrupt normal gene expression, ultimately driving various disease phenotypes. Consequently, there is a critical need in functional genomics and precision medicine to accurately decipher how genetic mutations influence gene regulation. Understanding these dynamics is essential for advancing disease diagnostics, prioritizing clinical variants, and developing targeted therapeutic interventions.

Despite this need, current methodologies face significant limitations in mapping these regulatory effects accurately. Traditional bulk-tissue gene expression models analyze heterogeneous mixtures of cells, which averages out distinct transcriptional signals across the sample. This lack of high-resolution data masks critical regulatory dynamics that occur exclusively within specific, individual cell types. Because many diseases originate from dysregulation in highly specialized cellular subpopulations, these bulk approaches fail to capture the true biological context of genetic mutations. As a result, researchers struggle to accurately prioritize noncoding variants or determine their functional consequences in the specific tissues where they exert pathogenic effects.

Technology

This computational framework utilizes a modular deep learning architecture to predict cell-type-specific gene expression directly from DNA sequence data. It employs a deep convolutional neural network to process sequences flanking transcription start sites, modeling spatial patterns of epigenetic markers like chromatin accessibility. These predicted features are integrated into cell-type-specific regularized linear regression models trained on single-cell or single-nucleus RNA sequencing data. This allows the system to estimate expression levels for thousands of genes across various primary cell types and calculate expression changes between reference and alternative alleles to assess genetic polymorphisms.

The technology is differentiated by overcoming the limitations of traditional bulk-tissue models, which lack the resolution to capture regulatory effects at the individual cell-type level. By leveraging a unique single-cell-trained framework, it achieves high-resolution predictions across over 100 primary human cell types, significantly outperforming whole-organ models. This precision enables the in silico screening of rare or unobserved noncoding variants, facilitating highly accurate variant prioritization for diagnostics. Furthermore, it empowers advanced therapeutic applications, such as guiding site-directed mutagenesis and ex vivo gene editing, by pinpointing exactly how genetic variations influence disease phenotypes in specific tissues.

Benefits

• Enables highly accurate, high-resolution prediction of gene expression for specific primary cell types directly from DNA sequences, overcoming the limitations of traditional bulk-tissue models.
• Allows rapid in silico assessment of how genetic variations, including rare and unobserved mutations, impact gene regulation and expression.
• Improves computational diagnostics by identifying misregulated genes and prioritizing disease-associated noncoding variants to inform early detection and personalized medicine.
• Facilitates the development of targeted gene-editing therapies (e.g., CRISPR) by predicting the precise effects of site-directed mutagenesis on specific cell types.
• Accelerates functional genomics research and guides experimental design by computationally screening variants prior to costly empirical validation.
• Features a flexible, modular architecture that allows independent updates to the epigenetic prediction models and the cell-type-specific expression models as new data becomes available.

Applications

• Genetic variant diagnostic screening
• Targeted sequencing kit development
• CRISPR gene editing design
• Personalized medicine treatment planning
• Therapeutic variant target prioritization

Additional info

This modular framework utilizes deep convolutional neural networks to model spatial epigenetic patterns from DNA sequences. These features are integrated into cell-type-specific regularized regression models, trained on single-cell RNA sequencing data, to predict high-resolution gene expression. By calculating log-fold changes between alleles, the system prioritizes genetic variants and facilitates site-directed mutagenesis. It enables interpreting functional consequences of genetic variation for precision medicine, diagnostics, and therapeutic research.