Back

More cells, more doublets in highly multiplexed single-cell data

Howitt, G.; Dixit, G.; Aharon, R.; Streeton-Cook, V.; Ling, L.; Hickey, P. F.; Amann-Zalcenstein, D.; Gubbels, L.; Shanthikumar, S.; Ranganathan, S.; Neeland, M.; Maksimovic, J.; Oshlack, A.

2024-10-04 bioinformatics
10.1101/2024.10.03.616596 bioRxiv
Show abstract

Withdrawal statementThe authors have withdrawn this manuscript. In our initial submission of this paper, we proposed a combinatoric model for the probability of finding doublets in sample-barcoded single-cell RNA-sequencing data. This model predicts a doublet rate higher than the rate included in the documentation for the 10X Flex protocol. This was motivated by our use of doublet-finding software, scDblFinder, on experimental data using the Flex protocol, which identified several times more doublets than we expected based on the documentation. Our original model produced better agreement with the results of scDblFinder in 9 of our own data sets as well as one public dataset made available by 10X. During revisions, however, we performed Monte Carlo simulations of doublet formation that were inconsistent with the results of our model, and much closer to the predictions from the documentation. This prompted us to reanalyse the assumptions of our model. A longer version of this withdrawal statement that explains in detail the errors in the model and includes updated figures is available at https://github.com/Oshlack/flex-doublets, but in brief, our initial assumptions over-predicted the doublet fraction, and our revised model is now consistent with both the Monte Carlo simulations and the predictions from the Flex documentation. We therefore speculate that the unexpectedly high doublet fractions observed in the experimental data might be due to some combination of experimental conditions and software performance. Therefore, the authors do not wish this work to be cited as reference for the project. If you have any questions, please ontact the corresponding author.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.