BABEL enables cross-modality translation between multiomic profiles at single-cell resolution.

Area of Science:

• Single-cell biology
• Computational biology
• Genomics

Background:

• Simultaneous multiomic profiling in single cells is a major challenge.
• Existing paired single-cell RNA sequencing (scRNA-seq) and scATAC-seq methods are complex and costly.
• There is a need for methods to computationally infer one modality from another.

Purpose of the Study:

• Introduce BABEL, a deep learning method for translating between single-cell transcriptome and chromatin accessibility profiles.
• Enable computational synthesis of paired multiomic data when only one modality is experimentally available.
• Demonstrate BABEL's accuracy, generalization, and applicability to new biological contexts.

Main Methods:

• Developed BABEL, an interoperable neural network model for cross-modality prediction.
• Trained BABEL on paired scRNA-seq and scATAC-seq datasets from human and mouse.
• Validated BABEL's predictive accuracy on independent datasets and novel cell types.

Main Results:

• BABEL accurately predicts gene expression from scATAC-seq and vice versa for individual cells.
• The method generalizes to new cell types and biological contexts, as shown in basal cell carcinoma (BCC) analysis.
• BABEL-generated expression data enabled fine-grained cell state classification comparable to experimental scRNA-seq.
• The model can integrate additional modalities like protein epitope profiling.

Conclusions:

• BABEL provides a powerful computational tool to overcome limitations in experimental multiomic profiling.
• It facilitates data exploration and hypothesis generation by enabling cross-modality analysis.
• This approach expands the utility of single-cell multiomic data in biological research.