Integrating single-cell multimodal epigenomic data using 1D convolutional neural networks.

Area of Science:

• Computational Biology
• Genomics
• Epigenetics

Background:

• Single-cell multimodal epigenomic profiling measures multiple histone modifications and chromatin accessibility simultaneously.
• Integrating these diverse epigenomic datasets is crucial for understanding cell-type variations.
• Existing integration methods are not optimized for the unique characteristics of multimodal epigenomic data.

Purpose of the Study:

• To develop a novel framework for integrating single-cell multimodal epigenomic data.
• To model this data as a multichannel sequential signal using convolutional variational autoencoders.
• To improve dimension reduction and batch correction for epigenomic datasets.

Main Methods:

• Developed ConvNet-VAEs, a framework utilizing 1D convolutional variational autoencoders (VAEs).
• Applied ConvNet-VAEs to nano-CUT&Tag and single-cell nanobody-tethered transposition sequencing data.
• Evaluated performance on juvenile mouse brain and human bone marrow datasets.

Main Results:

• ConvNet-VAEs demonstrated superior dimension reduction and batch correction compared to existing architectures.
• Performance advantages increased with the number of epigenomic modalities analyzed.
• Deeper convolutional architectures enhanced performance, while deeper fully connected architectures showed degradation.

Conclusions:

• ConvNet-VAEs offer a promising approach for integrating single-cell multimodal epigenomic data.
• Convolutional autoencoders are well-suited for current and future epigenomic datasets.
• The framework provides improved efficiency and performance in data integration.