DeconPeaker, a Deconvolution Model to Identify Cell Types Based on Chromatin Accessibility in ATAC-Seq Data of...

Understanding cell identity remains a challenge due to cellular heterogeneity and dynamic changes in chromatin architecture. Chromatin accessibility patterns are critical for gene regulation and disease progression. However, accurately identifying cell types in mixed samples is difficult. Prior research has shown that gene expression data alone is insufficient for precise cell-type deconvolution. Open chromatin data, such as from ATAC-seq, offers a more detailed view of regulatory elements. Yet, no prior work had resolved how to combine chromatin accessibility with gene expression for improved cell-type identification. This gap motivated the development of a new deconvolution method that leverages chromatin accessibility. Existing tools lack the precision needed for complex biological systems. A novel approach was required to address these limitations. This paper introduces DeconPeaker to fill this need.

The study aimed to develop a deconvolution tool that integrates chromatin accessibility and gene expression data to identify cell types in mixed samples. Chromatin accessibility is a key indicator of cell identity, but its use in deconvolution is limited. The researchers sought to improve accuracy by combining multiple data types. They hypothesized that chromatin accessibility provides more reliable signals than gene expression alone. The tool was designed to estimate cell-type proportions from ATAC-seq data. The goal was to test DeconPeaker's performance against existing methods. The researchers also wanted to apply it to a real-world disease model. The study aimed to demonstrate the tool's potential in clinical settings.

DeconPeaker was developed as a computational model using chromatin accessibility data from ATAC-seq. The method uses open chromatin peaks to infer cell-type composition in mixed samples. The model was trained on simulated datasets with known cell-type proportions. The researchers compared DeconPeaker to existing deconvolution tools using root-mean-square error (RMSE) and correlation coefficients. They evaluated performance using both simulated and real data from acute myeloid leukemia (AML). Chromatin accessibility and gene expression datasets were analyzed together. The model was tested for accuracy in identifying cell types and their proportions. The researchers validated the tool using AML chromatin data to assess its biological relevance.

DeconPeaker achieved the lowest average RMSE of 0.042 among tested deconvolution methods. The tool also showed the highest average correlation coefficient of 0.919 between predicted and true cell-type proportions. Chromatin accessibility data outperformed gene expression in cell-type identification. The model successfully identified unique cell types in AML chromatin data. DeconPeaker demonstrated superior accuracy compared to existing tools. The combination of chromatin accessibility and gene expression improved deconvolution performance. The model's performance was validated using both simulated and real datasets. The results suggest that chromatin accessibility is a more reliable marker for cell-type identification.

The authors propose that DeconPeaker is a powerful tool for cell-type deconvolution using chromatin accessibility data. The model's performance suggests it is more effective than existing methods. The study shows that chromatin accessibility provides better resolution than gene expression alone. DeconPeaker can be used to analyze mixed samples in complex biological systems. The tool's success in AML data indicates its potential in clinical applications. The researchers suggest that combining chromatin accessibility and gene expression improves accuracy. The model's availability on GitHub allows for further validation and use. The study concludes that DeconPeaker enhances the ability to identify cell types in heterogeneous samples.