Abstract
Single-cell RNA sequencing methods based on split-pool combinatorial barcoding enable high-throughput profiling, yet sample identity is often encoded during early barcoding steps rather than through the library index. Consequently, reads from multiple biological samples remain pooled, complicating per-sample analysis and selective extraction of samples of interest. Here, I present CapMux, a Snakemake-based pipeline for processing split-pool scRNA-seq data from raw sequencing files to sample-resolved outputs. CapMux supports workflows starting from either BCL files or FASTQ files and reconstructs sample identity by integrating sub-library index information with the experiment-specific barcoding plate layout. The pipeline was developed for the CapSeq method but is configurable for related scRNA-seq combinatorial barcoding designs through specification of barcode positions and experimental layout. In a controlled cell line mixing scRNA-seq experiment, CapMux resolved pooled data into outputs for each sample, enabling independent quality control summaries, mapping statistics, count matrices, and downstream visualizations. Runtime benchmarking indicated that secondary demultiplexing step added only a modest computational overhead. Together, these results show that CapMux provides a practical and adaptable framework for recovering sample-level resolution from split-pool scRNA-seq data.
1 Introduction
Single-cell sequencing methods have become central to modern molecular biology and biomedicine because they enable the resolution of cellular heterogeneity that is obscured in bulk measurements (; ; ). Reflecting this impact, single-cell sequencing was named “Method of the Year” in 2013 (). Since then, the field has expanded rapidly, from transcriptome profiling to a broad and growing repertoire of single-cell and spatial multi-omics approaches that integrate genomic, epigenomic, proteomic, spatial, and other molecular layers at single-cell resolution (; ).
Among these technologies, single-cell RNA sequencing (scRNA-seq) remains one of the most widely used modalities and is supported by a large ecosystem of commercial and open-source platforms (; ; ; ). Current scRNA-seq methods vary in throughput, transcript coverage, sensitivity, and experimental complexity, but high-throughput approaches are commonly based on either droplet-based compartmentalization or plate-based combinatorial indexing strategies (; ; ; ; ; ). In both cases, sequencing reads are associated with synthetic cell-identifying barcodes; however, the way sample identity is encoded and later recovered differs substantially between these two families of methods.
In conventional droplet-based methods, a sequencing library is typically prepared for a single experimental sample, and library indexes can therefore be used to separate libraries at an early stage of analysis (; ; ; ). By contrast, in plate-based combinatorial barcoding methods, sample identity is often introduced during the first barcoding round (e.g., reverse transcription), after which cells or nuclei are pooled and redistributed across subsequent rounds of barcoding (; ; ; ). As a result, sample identity cannot be resolved from the final sub-library index alone and instead must be reconstructed computationally by integrating sub-library index information with the internal barcode structure.
This design creates a practical bottleneck for downstream analysis. In many split-pool methods, raw sequencing data from multiple samples remain entangled until barcode parsing, alignment, and count-matrix generation have already progressed substantially. Consequently, extracting a single sample of interest, performing per-sample quality control, or reanalyzing only a subset of samples with alternative downstream pipelines becomes unnecessarily cumbersome. The same issue also complicates data reuse from public repositories, where users often must download and process sub-library FASTQ files containing reads from pooled samples even when only a single sample is of interest. Although several processing workflows exist for split-pool scRNA-seq data, they are typically method-specific (; ; ), which limits their use as flexible, user-configurable tools for early demultiplexing of sub-libraries into sample-resolved datasets prior to downstream analysis.
Here, I present a bioinformatic pipeline designed to address this gap by enabling early and flexible demultiplexing of split-pool scRNA-seq data at the raw-read level, prior to downstream analysis. The pipeline was initially developed for the CapSeq scRNA-seq method (), but it was intentionally designed to remain adaptable to future changes in barcode architecture and experimental layout. In addition to supporting CapSeq, the framework can be configured for other similar split-pool combinatorial barcoding designs by specifying barcode positions, barcode counts, unique molecular identifier (UMI) structure, and indexing scheme. To make the pipeline accessible, users provide sample metadata and plate layout information through a simple input file and specify experiment-specific parameters in a configuration file, thereby minimizing the need for custom scripting while supporting diverse experimental designs.
2 Results
2.1 Pipeline architecture and workflow
The CapMux pipeline can be divided into several sequential steps, as illustrated in Figure 1A. The selected execution regime depends on how the user populates the configuration file and the sample sheet. Specifically, the pipeline can be initiated either from raw Illumina BCL files (run_mode = “bcl”) or from already demultiplexed FASTQ files (run_mode = “fastq”). In the main workflow, in which libraries are further demultiplexed according to the BC1 barcoding plate, the user selects demux_by = “bc1” and provides plate layout in the sample sheet that matches the experimental plate design. In contrast, when the experiment does not contain BC1-defined samples, the user can choose the standard demultiplexing mode, demux_by = “index”, which performs demultiplexing based solely on the sample index sequence.
FIGURE 1
In split-pool scRNA-seq experiments, the cell barcode (CB) typically consists of multiple barcode segments (for example, BC1-BC2-BC3), which may differ in length and may be separated by linker sequences (Figure 1C). To accommodate this variability, the configuration file allows the user to define the CB structure explicitly by specifying the length of each individual barcode segment. At present, the pipeline supports construction of CBs composed of one to three segments and, when appropriate, one additional barcode segment located in the index read (i7 or i5). The user can also define the unique molecular identifier (UMI) length and specify whether the UMI is contiguous or split into two parts, as in the CapSeq barcoding scheme (Figure 1C). Index files containing barcode sequences must be placed in the assets/barcodes directory, with the barcode sequences in the BC1 barcode index file ordered to match their positions on the physical BC1 barcoding plate, from A1 to H12. Taken together, these settings enable flexible adaptation of the pipeline to different experimental designs while maintaining a standardized processing framework.
In the main CapMux execution regime (Figure 1A, blue), the workflow consists of the following key steps. First, bcl2fastq performs primary demultiplexing by converting raw sequencing data into standard FASTQ files based on the index read. Second, bc1 demux carries out secondary demultiplexing, splitting index-based FASTQ files into sample-specific FASTQ files according to the user-defined BC1 plate layout. Third, merge bc assembles the barcode read by reconstructing the complete CB together with the UMI sequence while removing linker sequences. Next, Cutadapt trims the template switching oligo (TSO) sequence from complementary DNA (cDNA) reads. FastQC is then used for quality control of the FASTQ files, followed by STARsolo, which aligns reads and generates a count matrix for each sample. Finally, MultiQC compiles a structured quality control summary for all demultiplexed samples. Completion of the pipeline yields a structured set of output files within the results directory (Figure 1B). Among these, the principal outputs are demultiplexed FASTQ files, aligned BAM files, gene expression count matrices, and quality control reports generated for each individual sample (Figure 2B).
FIGURE 2
2.2 Sample demultiplexing: resolving sample identity after combinatorial barcoding
As illustrated in Figure 2, sample identity is introduced during the wet-lab procedure through the BC1 barcoding plate, but this positional information is lost after pooling and multiple rounds of combinatorial barcoding. CapMux resolves this by applying a secondary demultiplexing step before alignment. In this step, each group of FASTQ files generated for a single index during primary demultiplexing is processed independently and handled in parallel. For each group, the pipeline extracts the BC1 sequence from the user-defined barcode structure, matches it to the sample layout on the BC1 plate derived from the sample sheet, and writes reads into multiple sample-specific intermediate FASTQ files. BC1 matching can be performed either by exact whitelist matching or, when mismatch correction is enabled through the barcode.allow_mismatches parameter, by considering BC1 sequences that differ by a single nucleotide substitution from a whitelist barcode. Reads are assigned to a sample only when the observed or corrected BC1 sequence maps uniquely to one whitelist barcode, while unmatched BC1 sequences are written to undetermined output files. At the same time, the pipeline reconstructs a barcode read by concatenating the configured barcode segments, index sequence, and UMI sequence, thereby generating the paired barcode and cDNA FASTQ outputs required for downstream mapping. After all index-based FASTQ files have been processed, intermediate files belonging to the same sample are merged and compressed in parallel to produce final per-sample FASTQ files for subsequent analysis.
2.3 CapMux showcase in a cell line mixing experiment
To demonstrate CapMux in a controlled setting, a CapSeq scRNA-seq experiment was performed using 3 mouse cell lines (Figure 3A). Each cell line was distributed across four BC1 plate columns: Neuro-2a (N2a) cells in columns 1–4, NIH/3T3 cells in columns 5–8, and mouse embryonic stem cells (mESCs) in columns 9–12.
FIGURE 3
Following sequencing, raw BCL files were processed with CapMux using two execution regimes, both with run_mode = “bcl” but differing in the demux_by parameter (Figure 3B). In the first regime, demux_by = “index”, only primary index-based demultiplexing was performed, resulting in a pooled output. In the second regime, demux_by = “bc1”, the pipeline performed secondary demultiplexing and generated sample-resolved outputs.
Comparison of the resulting quality control summaries and UMAP visualizations (Figure 4) highlights the main advantage of secondary demultiplexing with demux_by = “bc1”: sequencing, alignment, and downstream summary statistics were retained separately for each individual sample, whereas under demux_by = “index” these data remained combined and sample-specific differences were masked. In the alignment summaries (Figure 4A) and feature assignment categories (Figure 4B), the demux_by = “index” regime reported only a single combined profile for the pooled dataset, whereas the demux_by = “bc1” regime resolved the data into separate sample-specific profiles for N2a, NIH/3T3, and mESCs. Likewise, in the STARsolo summary tables (Figure 4C), the demux_by = “index” regime produced a single aggregated output, whereas demux_by = “bc1” produced three separate outputs corresponding to the 3 cell lines. This enabled direct comparison of sample-specific metrics, including the number of recovered cells, reads per cell, number of detected genes, and other summary metrics.
FIGURE 4
The same pattern was also evident in the UMAP projections (Figure 4D). Under demux_by = “index”, all cells remained combined within a single pooled output, whereas secondary demultiplexing with demux_by = “bc1” separated the experiment into three distinct sample-resolved datasets corresponding to the original BC1 plate layout. Together, these results show that CapMux recovers sample-level resolution from pooled CapSeq data, enabling direct comparison of sample-specific differences in read distribution, mapping statistics, feature assignment, and other output metrics in downstream analysis.
2.4 Runtime comparison across execution regimes
Workflow wall-clock runtime, measured after prior software environment preparation, differed only modestly across the tested CapMux execution regimes (Table 1; Figure 1A). Under identical input data, pre-created software environments, and fixed high-performance computing (HPC) resource conditions, execution regime A (run_mode = “bcl”, demux_by = “bc1”), which includes the additional secondary demultiplexing step, required only a small increase in runtime compared with execution regime B (run_mode = “bcl”, demux_by = “index”), in which only primary demultiplexing based on the index read is performed. In this dataset, which comprised three samples, the inclusion of secondary demultiplexing increased the mean execution time by approximately 4 min. As expected, the fastest configuration was execution regime C (run_mode = “fastq”, demux_by = “bc1”), because this regime starts from index-demultiplexed FASTQ files and therefore bypasses primary BCL-to-FASTQ conversion and index-based demultiplexing. Overall, these results suggest that the additional secondary demultiplexing step adds only a modest runtime overhead.
TABLE 1
| Execution regime | run_mode | demux_by | Runtime |
|---|---|---|---|
| A | bcl | bc1 | 44 min 51 s 1min 56 s |
| B | bcl | index | 40 min 46 s 2 min 14 s |
| C | fastq | bc1 | 35 min 58 s 50 s |
Comparison of workflow wall-clock runtime across different CapMux execution regimes. Reported values represent the mean SD from three runs performed under identical input and HPC resource settings.
To further evaluate performance in a larger sample-multiplexed setting, CapMux was also applied to a previously generated CapSeq dataset () containing 12 BC1-defined samples distributed across ten sub-libraries. This larger dataset contained approximately 1 billion reads, compared with approximately 100 million reads in the smaller dataset, corresponding to a 10 times larger input. The same HPC resource conditions were used as for the smaller dataset, and execution regime A (run_mode = “bcl”, demux_by = “bc1”) and execution regime B (run_mode = “bcl”, demux_by = “index”) were evaluated. This analysis showed that the same demultiplexing strategy could be readily scaled without changes to the pipeline logic. As expected, runtime increased with the larger input size: regime A required 388 min 19 s 7 min 35 s, whereas regime B required 313 m 58 s 20 m 2 s. Thus, the inclusion of BC1-based secondary demultiplexing increased mean runtime by approximately 24%, while enabling the generation of sample-resolved outputs for all 12 samples.
3 Discussion
Split-pool combinatorial barcoding methods have become increasingly important in scRNA-seq because they enable profiling of very large numbers of cells in a single experiment while avoiding some of the physical constraints of droplet-based methods. However, the same experimental logic that makes these methods scalable also creates a downstream bioinformatics challenge: samples are often pooled early, whereas sample identity is encoded within the internal barcode structure rather than in the final sequencing index alone. As a result, sample-level resolution is often recovered only late in processing, after substantial computational work has already been completed. The results presented here show that CapMux addresses this bottleneck by enabling early reconstruction of sample identity and generating sample-resolved outputs before alignment and count-matrix generation.
CapMux operates at a different stage of the single-cell analysis workflow than downstream analysis tools such as Seurat () or Scanpy (). These tools are typically used after alignment and count-matrix generation, where sample identity can be inferred, for example, through hashing-based approaches or added manually by assigning cell barcodes to sample labels at the count-matrix stage. In contrast, CapMux operates upstream of these analyses, directly at the BCL or FASTQ processing stage. By combining the sub-library index with the internal barcode structure and experiment-specific plate layout, CapMux generates sample-resolved FASTQ files before alignment and quantification. This automates processing directly from raw sequencing data and allows users to extract samples of interest, perform per-sample quality control, or reanalyze only a subset of samples with selected downstream pipelines.
A central idea of this work is that early secondary demultiplexing improves the interpretability of split-pool scRNA-seq datasets. In the cell line mixing experiment, secondary demultiplexing separated pooled data into outputs corresponding to the original samples, thereby revealing sample-specific differences that were obscured when only primary demultiplexing based on the index read was used. This was evident across several levels of analysis, including summary metrics, alignment statistics, feature assignment profiles, and UMAP visualizations. That distinction is practically important, because sample-resolved outputs represent a more natural unit for biological interpretation, troubleshooting, and comparison across experimental conditions.
Although CapMux was developed for the CapSeq method (), its design was intentionally made more general. Rather than hard-coding a single barcode architecture, the pipeline allows users to define barcode segments and UMI structure. This makes the framework potentially applicable beyond CapSeq to other split-pool methods that follow a similar barcoding logic. Such flexibility may be particularly useful for the development of new or customized split-pool scRNA-seq methods, as existing method-specific pipelines are optimized for a fixed barcode architecture. For example, commercial workflows such as ScaleRna () provide end-to-end processing for ScaleBio single-cell platforms, including demultiplexing, alignment, count-matrix generation, and quality-control reporting. Such workflows are valuable for processing data generated with their intended protocols, but newly developed or modified split-pool designs may require substantial custom scripting or dedicated preprocessing workflows before sample-resolved analysis can be performed. CapMux addresses this by allowing users to define experiment-specific barcode positions, UMI structure, indexing scheme, and sample layout information. Thus, CapMux provides a customizable upstream preprocessing framework that may reduce the need to build a new demultiplexing pipeline for each modified or newly developed split-pool barcoding design.
An additional practical observation is that the benefit of secondary demultiplexing was achieved with only a modest runtime penalty. Under the tested HPC conditions, inclusion of the extra demultiplexing step increased mean runtime by 10% in the smaller dataset, whereas in the dataset that was 10 times larger, mean runtime increased by approximately 24%. Although absolute runtimes will depend on dataset size, hardware, storage performance, and experimental design, the results suggest that early reconstruction of sample identity has a manageable impact on overall workflow runtime.
Taken together, CapMux provides a practical solution for processing split-pool scRNA-seq raw data into sample-resolved FASTQ files and downstream outputs. Future work should test CapMux across a wider range of split-pool methods, expand support for additional barcode architectures, incorporate further functionality such as alternative aligners or quantification tools, expanded quality control reporting, and tighter integration with downstream single-cell analysis workflows. Because the pipeline is implemented in a modular Snakemake framework, these developments should be straightforward to integrate, which may help CapMux remain useful as single-cell methods based on combinatorial barcoding continue to gain popularity.
4 Materials and methods
4.1 Data processing software
The CapMux workflow is implemented in Snakemake (v9.16.3) () and executed with per-rule Conda environments to ensure dependency isolation and reproducibility. Illumina base call conversion and initial demultiplexing were performed with bcl2fastq (v2.19.0). Tabular data processing was handled using custom Python scripts built on pandas (v2.2.3) and openpyxl (v3.1.5). TSO sequence trimming was performed with Cutadapt (v5.0) (). Read alignment and single-cell quantification were carried out with STAR (v2.7.11b) (). Read-level quality control was generated with FastQC (v0.12.1) (), and run-level summaries were aggregated into a single report using MultiQC (v1.28) (). Additional preprocessing in custom scripts relied on standard Unix command-line utilities, including GNU coreutils (v9.5), GNU awk/gawk (v5.3.1), gzip (v1.14), and pigz (v2.8). Custom scripts were run under Python (v3.13.12).
4.2 Runtime benchmarking
Pipeline performance was evaluated using end-to-end wall-clock runtime measured at the level of the Slurm batch step, after prior preparation of all required software environments. Snakemake Conda environments were created in a separate preparatory step before benchmarking, and environment creation or package download time was excluded from runtime measurements. Runtime was therefore defined as the elapsed time from initiation of the batch script on the allocated compute node until script termination during workflow execution. For each execution regime, the same input dataset was processed under identical HPC resource allocation conditions (16 CPU cores, 64 GB RAM). Runtime was obtained from Slurm accounting (sacct).
4.3 Cell lines
NIH/3T3 (ATCC, CRL-1658) and Neuro-2a (ATCC, CCL-131) cells were grown in Dulbecco’s Modified Eagle Medium (DMEM, Gibco, 61965026) supplemented with 10% fetal bovine serum (FBS) and 1 penicillin-streptomycin (PS; Gibco, 15140122). Mouse embryonic stem cells (ATCC, CRL-1821, E14TG2a) were cultivated on dishes coated with 0.15% gelatin in DMEM supplemented with 15% FBS, 1 PS, 0.1 mM sodium pyruvate (Gibco, 11360039), 0.1 mM -mercaptoethanol (Gibco, 31350010), 1 mM L-alanyl-L-glutamine (Gibco), 1 Non-Essential Amino Acids (Gibco, 11140050), 1103 U/mL mouse leukemia inhibitory factor (Millipore), 3 CHIR99021 (Sigma) and 1 PD0325901 (Sigma). All cell lines were grown at 5% and at 37 °C. Cells were collected at roughly 1106cells/mL, transferred into 15 mL conical tubes, pelleted by centrifugation (100–300 g for 5 min–7 min), and washed twice with ice-cold 1 Dulbecco’s Phosphate-Buffered Saline (DPBS; Gibco 14190144). Cell number and viability were assessed with a hemocytometer using 0.2% trypan blue staining.
4.4 Cell encapsulation and scRNA-seq experiment
Cell encapsulation was performed as previously described . Briefly, cells at 3 M/mL concentration in 1 DPBS were mixed with 30% (w/w) dextran (Darna Bio, DB-DEX-2001) at equal volumes, achieving a final concentration of 1.5 M/mL. The stock 10% (w/w) gelatin methacryloyl solution (GelMA; Darna Bio, DB-GMA-2001) was pre-warmed and diluted with 1 DPBS to the 4% (w/v) working solution. Cell encapsulation was carried out using a 40 -deep microfluidic device, with flow rates set to 100 for the cell suspension containing dextran, 200 for the GelMA working solution, and 700 for a surfactant (Darna Bio, DB-SRF-2001). Encapsulations were performed at 26 °C. The remaining protocol steps, including the scRNA-seq experiment, were conducted following the CapSeq procedure () without the hairpin ligation step. Sequencing was performed on the Illumina NextSeq 2000 platform using the following run settings: R1, 43 cycles; i7, 6 cycles; and R2, 89 cycles.
4.5 Raw sequencing data processing
To generate cell gene count matrices, the CapMux snakemake pipeline was run under {run_mode: “bcl”, demux_by: “bc1”}, {run_mode: “bcl”, demux_by: “index”}, and {run_mode: “fastq”, demux_by: “bc1”} execution regimes. All other parameters were identical across runs. Barcode segments were defined as follows: bc1:{start: 32, len: 8}, bc2:{start: 18, len: 10}, bc3:{start: 5, len: 8}, umi1:{start: 1, len: 4}, umi2:{start: 40, len: 4}. Reads were aligned to the mouse GRCm39 genome (GENCODE M30).
4.6 UMAP construction from scRNA-seq data
Single-cell RNA sequencing data was analyzed in Python using Scanpy (v1.11.4) (). Low-quality cells were filtered out from the cell gene count matrices by requiring a minimum of 1,000 UMIs/cell and excluding cells with mitochondrial gene count fractions above 20%. UMAP embeddings were generated from normalized and filtered gene expression matrices. Counts were normalized to 10,000 per cell, after which mitochondrial and ribosomal genes were excluded. Additional gene-level filtering was applied to reduce noise: genes detected in fewer than 5 cells and with expression below 10 counts were removed. The data was then log-transformed. Highly variable genes were identified using scanpy.pp.highly_variable_genes (flavor = “seurat_v3”), and the top 150 genes were selected. These genes were z-score scaled, followed by principal component analysis. A k-nearest neighbor graph was constructed using scanpy.pp.neighbors (n_neighbors = 100), and UMAP embeddings were computed with scanpy.tl.umap (min_dist = 0.3).
Statements
Data availability statement
The sequencing data generated in this study are publicly available in the Gene Expression Omnibus (GEO) repository under accession number GSE324486. The CapMux pipeline is available at https://github.com/Boyoron/CapMux.
Ethics statement
Ethical approval was not required for the studies on animals in accordance with the local legislation and institutional requirements because only commercially available established cell lines were used.
Author contributions
DB: Software, Visualization, Conceptualization, Investigation, Validation, Supervision, Formal Analysis, Methodology, Writing – original draft, Project administration, Writing – review and editing, Data curation.
Funding
The author(s) declared that financial support was received for this work and/or its publication. This research was funded by Project “Mission-driven Implementation of Science and Innovation Programmes” (No. 02-002-P-0001), which is funded by the Economic Revitalization and Resilience Enhancement Plan “New Generation Lithuania”.
Acknowledgments
I am sincerely grateful to Linas Mazutis, Simonas Juzenas, Juozapas Ivanauskas, and Simonas Norvaisis for their valuable discussions and constructive feedback.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was used in the creation of this manuscript. The author declares that Generative AI was used in the preparation of this manuscript. ChatGPT (OpenAI, GPT-5) was used solely for language editing to improve grammar, clarity, and wording. All scientific content, analysis, and interpretation were produced and verified by the author.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
1
BaronasD.NorvaisisS.ZvirblyteJ.LeonavicieneG.MikulenaiteV.GodaK.et al (2026). High-throughput single-cell omics using semipermeable capsules. Science391, 1138–1145. 10.1126/science.ady7227
2
BaysoyA.BaiZ.SatijaR.FanR. (2023). The technological landscape and applications of single-cell multi-omics. Nat. Rev. Mol. Cell Biol.24, 695–713. 10.1038/s41580-023-00615-w
3
CaoJ.PackerJ. S.RamaniV.CusanovichD. A.HuynhC.DazaR.et al (2017). Comprehensive single-cell transcriptional profiling of a multicellular organism. Science357, 661–667. 10.1126/science.aam8940
4
CaoJ.SpielmannM.QiuX.HuangX.IbrahimD. M.HillA. J.et al (2019). The single-cell transcriptional landscape of mammalian organogenesis. Nature566, 496–502. 10.1038/s41586-019-0969-x
5
ClarkI. C.FontanezK. M.MeltzerR. H.XueY.HayfordC.May-ZhangA.et al (2023). Microfluidics-free single-cell genomics with templated emulsification. Nat. Biotechnol.41, 1–10doi. 10.1038/s41587-023-01685-z
6
ConteM. I.Fuentes-TrilloA.CondeC. D. (2024). Opportunities and tradeoffs in single-cell transcriptomic technologies. Trends Genet.40, 83–93. 10.1016/j.tig.2023.10.003
7
De RopF. V.IsmailJ. N.Bravo González-BlasC.HulselmansG. J.FlerinC. C.JanssensJ.et al (2022). Hydrop enables droplet-based single-cell ATAC-seq and single-cell RNA-seq using dissolvable hydrogel beads. eLife11, e73971. 10.7554/eLife.73971
8
De SimoneM.HooverJ.LauJ.BennettH. M.WuB.ChenC.et al (2025). A comprehensive analysis framework for evaluating commercial single-cell RNA sequencing technologies. Nucleic Acids Res.53, gkae1186. 10.1093/nar/gkae1186
9
EwelsP.MagnussonM.LundinS.KällerM. (2016). MultiQC: summarize analysis results for multiple tools and samples in a single report. Bioinformatics32, 3047–3048. 10.1093/bioinformatics/btw354
10
FastQC (2010). Babraham bioinformatics - FastQC A quality control tool for high throughput sequence data. Available online at: https://www.bioinformatics.babraham.ac.uk/projects/fastqc/ (Accessed March 15, 2026).
11
JongheJ. D.OpzoomerJ. W.Vilas-ZornozaA.NilgesB. S.CraneP.VicariM.et al (2024). scTrends: a living review of commercial single-cell and spatial ’omic technologies. Cell Genomics4. 10.1016/j.xgen.2024.100723
12
KaminowB.YunusovD.DobinA. (2021). Starsolo: Accurate, Fast and Versatile mapping/quantification of single-cell and single-nucleus RNA-seq Data. 10.1101/2021.05.05.442755
13
KleinA. M.MazutisL.AkartunaI.TallapragadaN.VeresA.LiV.et al (2015). Droplet barcoding for single-cell transcriptomics applied to embryonic stem cells. Cell161, 1187–1201. 10.1016/j.cell.2015.04.044
14
KuijpersL.HornungB.van den Hout - van VroonhovenM. C. G. N.van IjckenW. F. J.GrosveldF.MulugetaE. (2024). Split pool Ligation-based single-cell transcriptome sequencing (SPLiT-seq) data processing pipeline comparison. BMC Genomics25, 361. 10.1186/s12864-024-10285-3
15
LiH.HumphreysB. D. (2022). Mouse kidney nuclear isolation and library preparation for single-cell combinatorial indexing RNA sequencing. Star. Protoc.3, 101904. 10.1016/j.xpro.2022.101904
16
MartinM. (2011). Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet.J.17, 10–12. 10.14806/ej.17.1.200
17
MölderF.JablonskiK. P.LetcherB.HallM. B.van DykenP. C.Tomkins-TinchC. H.et al (2025). Sustainable Data Analysis with Snakemake. 10.12688/f1000research.29032.3
18
Nature Methods (2013). Method of the year 2013. Nat. Methods11, 1. 10.1038/nmeth.2801
19
RosenbergA. B.RocoC. M.MuscatR. A.KuchinaA.SampleP.YaoZ.et al (2018). Single-cell profiling of the developing mouse brain and spinal cord with split-pool barcoding. Science360, 176–182. 10.1126/science.aam8999
20
SantP.RippeK.MallmJ.-P. (2023). Approaches for single-cell RNA sequencing across tissues and cell types. Transcription14, 127–145. 10.1080/21541264.2023.2200721
21
ScaleRna (2022). ScaleRna docs. Available online at: https://scalebio.github.io/ScaleRna-docs/ (Accessed March 15, 2026).
22
StuartT.ButlerA.HoffmanP.HafemeisterC.PapalexiE.MauckW. M.et al (2019). Comprehensive integration of single-cell data. Cell177, 1888–1902.e21. 10.1016/j.cell.2019.05.031
23
TranV.PapalexiE.SchroederS.KimG.SapreA.PangalloJ.et al (2022). High Sensitivity Single Cell RNA Sequencing with Split Pool Barcoding. 10.1101/2022.08.27.505512
24
VandereykenK.SifrimA.ThienpontB.VoetT. (2023). Methods and applications for single-cell and spatial multi-omics. Nat. Rev. Genet.24, 494–515. 10.1038/s41576-023-00580-2
25
WolfF. A.AngererP.TheisF. J. (2018). SCANPY: large-scale single-cell gene expression data analysis. Genome Biol.19, 15. 10.1186/s13059-017-1382-0
26
XieY.ChenH.ChellamuthuV. R.LajamA. b. M.AlbaniS.LowA. H. L.et al (2024). Comparative analysis of single-cell RNA sequencing methods with and without sample multiplexing. Int. J. Mol. Sci.25, 3828. 10.3390/ijms25073828
27
ZhengG. X. Y.TerryJ. M.BelgraderP.RyvkinP.BentZ. W.WilsonR.et al (2017). Massively parallel digital transcriptional profiling of single cells. Nat. Commun.8, 14049. 10.1038/ncomms14049
28
ZilionisR.NainysJ.VeresA.SavovaV.ZemmourD.KleinA. M.et al (2017). Single-cell barcoding and sequencing using droplet microfluidics. Nat. Protoc.12, 44–73. 10.1038/nprot.2016.154
Summary
Keywords
CapSeq, sample demultiplexing, sequencing data processing, single-cell RNA sequencing, split-pool combinatorial barcoding
Citation
Baronas D (2026) CapMux: a Snakemake pipeline for early demultiplexing of split-pool scRNA-seq data into sample-resolved outputs. Front. Bioinform. 6:1846065. doi: 10.3389/fbinf.2026.1846065
Received
02 April 2026
Revised
11 May 2026
Accepted
25 May 2026
Published
12 June 2026
Volume
6 - 2026
Edited by
Maciej Pietrzak, The Ohio State University, United States
Updates
Copyright
© 2026 Baronas.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Denis Baronas, denis.baronas@gmc.vu.lt
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.