Do we really know the impact of multiple single-cell data types?

We often assume that single-cell sequencing gives us a complete picture of cellular diversity, but actually, we already saw in previous publications, like here and here, that one protocol is not really enough to capture the full complexity of a tissue. So, what about integrating more data types? This is the question we try to answer in our recent study now on biorXiv.

Complex tissues, like the kidney cortex, are made up of many specialized cells, but no single sequencing method can fully capture all of them. That’s why combining different single-cell data types, such as RNA and ATAC, is essential to achieve a more complete view of their identity and function.

To address this, we used the anchor-based integration framework (well explained by Arguelaguet et al., 2021) and benchmarked three main types of single-cell data integration by using our new analytical framework scOMM:

• Horizontal integration: Combining datasets from the same modality (e.g., multiple RNA-seq protocols).

• Vertical integration: Merging data from multiomic assays (e.g., RNA+ATAC from the same nucleus).

• Diagonal/Mosaic integration: Integrating different single-modality assays (e.g., RNA from one dataset with ATAC from another).
  

What We Found

• Integrating multiple RNA-seq data types (3’, 5’, snRNA) with snATAC-seq significantly improves cell (sub)type resolution and marker identification.

• Not all cell types benefit equally, but certain populations, especially continuous states, are much better resolved with multimodal integration.

• Rare and clinically relevant populations, such as Norn cells (EPO-producing fibroblasts) and WFDC2+ thick ascending limb cells, were only detected when combining multiple data types, highlighting the power of multimodal approaches in discovering new biology.

Benchmarking summary.

Why This Matters

If we want to truly understand cellular complexity, we need to integrate data across platforms—only then can we uncover the full landscape of health and disease.

Please, read our preprint if you want to know more details about our approach! Tired of reading?? Ok, then listen to the podcast of our study generated by NotebookLM integrated in google.