# The impact of ambient contamination on demultiplexing methods for single-nucleus multiome experiments

## TL;DR

This paper studies how ambient contamination affects the accuracy of demultiplexing methods in single-nucleus multiome experiments and introduces a new tool and metric to improve this.

## Contribution

The paper introduces ambisim, a genotype-aware simulator, and a new metric called variant consistency to better model ambient contamination in demultiplexing.

## Key findings

- Demultiplexing methods are variably impacted by ambient contamination in both RNA and ATAC modalities.
- Concordance between demultiplexing methods is highly variable in joint snRNA/snATAC datasets.
- The new variant consistency metric correlates with ambient molecule fractions in singlets.

## Abstract

Sample multiplexing has become an increasingly common design choice in droplet-based single-nucleus multi-omic sequencing experiments to reduce costs and remove technical variation. Genotype-based demultiplexing is one popular class of methods that was originally developed for single-cell RNA-seq, but has not been rigorously benchmarked in other assays, such as snATAC-seq and joint snRNA/snATAC assays, especially in the context of variable ambient RNA/DNA contamination. To address this, we develop ambisim, a genotype-aware read-level simulator that can flexibly control ambient molecule proportions and generate realistic joint snRNA/snATAC data. We use ambisim to evaluate demultiplexing methods across several important parameters: doublet rate, number of multiplexed donors, and coverage levels. Our simulations reveal that methods are variably impacted by ambient contamination in both modalities. We then applied the demultiplexing methods to two joint snRNA/snATAC datasets and found highly variable concordance between methods in both modalities. Finally, we develop a new metric, variant consistency, which we show is correlated with cell-level ambient molecule fractions in singlets. Applying our metric to two multiplexed joint snRNA/snATAC datasets reveals variable ambient contamination across experiments and modalities. We conclude that improved modelling of ambient material in demultiplexing algorithms will increase both sensitivity and specificity.