Combining Results

After you have run each of the Demultiplexing and Doublet Detecting softwares you would like, it is helpful to convert them to similar nomenclature and combine the results into a single dataframe. In addition, we have found it helpful to generate summaries of each of the combinations of softwares identified. To help streamline this process, we have provided a script that will easily integrate all the softwares you have run into a single dataframe and can do the following:

1. Generate a dataframe that has all the software assignments per droplet in the pool

• A tab-separated dataframe with the droplet singlet-doublet classification and the individual assignment (for demultiplexing softwares) per droplet

1. Generate an upset plot that shows the droplet classificaitons by each software and the final classifications

2. Generate a droplet type summary file

• Provides the number of droplets classified for each combination of droplet classifications by each software

1. Generate demultiplexing individual assignment summary file

• Provides the number of droplets classified for each combination of individual assignment droplet classifications by each software

1. If individuals have not been assigned to each cluster for reference-free demultiplexing softwares, will create a common assignment across all demultiplexing softwares for easy comparison

2. Combined final droplet assignment from all softwares included

• Uses one of four intersectional methods to combine software assignments together into a single combined assignment per barcode

and to generate a summary file for all the software combinations and if you ran demultiplexing softwares, it will also generate a demultiplexing summary file for the individual and cluster assignments from the demultiplexing softwares.

Data

In order to use our script to combine the results from the various demultiplexing and doublet detecting softwares, you need the following:

Required

• Output directory ($OUTDIR)

• Path to results of each of the softwares you would like to merge into a single dataframe.

  • You need to provide the path to at least one software result, otherwise, it will not run.

Merging Results with Combine_Results.R

The script has multiple options to provide the paths to each of the software results you would like to run. To see each of the options, simply run:

singularity exec Demuxafy.sif Combine_Results.R -h

Providing the possible parameter options:

usage: /directflow/SCCGGroupShare/projects/DrewNeavin/Demultiplex_Benchmark/Demultiplexing_Doublet_Detecting_Docs/scripts/Combine_Results.R
      [-h] -o OUT [-d DEMUXLET] [-f FREEMUXLET] [-g FREEMUXLET_ASSIGNMENTS]
      [-a FREEMUXLET_CORRELATION_LIMIT] [-s SCSPLIT] [-w SCSPLIT_ASSIGNMENTS]
      [-j SCSPLIT_CORRELATION_LIMIT] [-u SOUPORCELL]
      [-x SOUPORCELL_ASSIGNMENTS] [-k SOUPORCELL_CORRELATION_LIMIT]
      [-v VIREO] [-e DOUBLETDECON] [-t DOUBLETDETECTION] [-i DOUBLETFINDER]
      [-n SCDBLFINDER] [-c SCDS] [-r SCRUBLET] [-l SOLO] [-b REF]
      [-p PCT_AGREEMENT] [-m METHOD]

optional arguments:
  -h, --help            show this help message and exit
  -o OUT, --out OUT     The file where results will be saved
  -d DEMUXLET, --demuxlet DEMUXLET
                        Path to demuxlet results. Only use this option if you
                        want to include the demuxlet results.
  -f FREEMUXLET, --freemuxlet FREEMUXLET
                        Path to freemuxlet results. Only use this option if
                        you want to include the freemuxlet results.
  -g FREEMUXLET_ASSIGNMENTS, --freemuxlet_assignments FREEMUXLET_ASSIGNMENTS
                        Path to freemuxlet cluster-to-individual assignments.
                        Only use this option if have used reference SNP
                        genotypes to assign individuals to clusters for the
                        freemuxlet results.
  -a FREEMUXLET_CORRELATION_LIMIT, --freemuxlet_correlation_limit FREEMUXLET_CORRELATION_LIMIT
                        The minimum correlation between the cluster and the
                        individual SNP genotypes which should be considered as
                        a valid assignment. If you want no limit, use 0.
                        Default is 0.7.
  -s SCSPLIT, --scSplit SCSPLIT
                        Path to scSplit results. Only use this option if you
                        want to include the scSplit results.
  -w SCSPLIT_ASSIGNMENTS, --scSplit_assignments SCSPLIT_ASSIGNMENTS
                        Path to scSplit cluster-to-individual assignments.
                        Only use this option if you have used reference SNP
                        genotypes to assign individuals to clusters for the
                        scSplit results.
  -j SCSPLIT_CORRELATION_LIMIT, --scSplit_correlation_limit SCSPLIT_CORRELATION_LIMIT
                        The minimum correlation between the cluster and the
                        individual SNP genotypes which should be considered as
                        a valid assignment. If you want no limit, use 0.
                        Default is 0.7.
  -u SOUPORCELL, --souporcell SOUPORCELL
                        Path to souporcell results. Only use this option if
                        you want to include the souporcell results.
  -x SOUPORCELL_ASSIGNMENTS, --souporcell_assignments SOUPORCELL_ASSIGNMENTS
                        Path to souporcell cluster-to-individual assignments.
                        Only use this option if you have used reference SNP
                        genotypes to assign individuals to clusters for the
                        souporcell results.
  -k SOUPORCELL_CORRELATION_LIMIT, --souporcell_correlation_limit SOUPORCELL_CORRELATION_LIMIT
                        The minimum correlation between the cluster and the
                        individual SNP genotypes which should be considered as
                        a valid assignment. If you want no limit, use 0.
                        Default is 0.7.
  -v VIREO, --vireo VIREO
                        Path to vireo results. Only use this option if you
                        want to include the vireo results.
  -e DOUBLETDECON, --DoubletDecon DOUBLETDECON
                        Path to DoubletDecon results. Only use this option if
                        you want to include the DoubletDecon results.
  -t DOUBLETDETECTION, --DoubletDetection DOUBLETDETECTION
                        Path to DoubletDetection results. Only use this option
                        if you want to include the DoubletDetection results.
  -i DOUBLETFINDER, --DoubletFinder DOUBLETFINDER
                        Path to DoubletFinder results. Only use this option if
                        you want to include the DoubletFinder results.
  -n SCDBLFINDER, --scDblFinder SCDBLFINDER
                        Path to scDblFinder results. Only use this option if
                        you want to include the scDblFinder results.
  -c SCDS, --scds SCDS  Path to scds results. Only use this option if you want
                        to include the scds results.
  -r SCRUBLET, --scrublet SCRUBLET
                        Path to scrublet results. Only use this option if you
                        want to include the scrublet results.
  -l SOLO, --solo SOLO  Path to solo results. Only use this option if you want
                        to include the solo results.
  -b REF, --ref REF     Which demultiplexing software to use as a reference
                        for individuals when you do not have assignment key
                        for all demultiplexing method. Options are 'Demuxlet',
                        'Freemuxlet', 'scSplit', 'Souporcell' and 'Vireo'. If
                        blank when assignment keys are missing, default
                        softwares to use if present are Vireo, then Demuxlet,
                        then Freemuxlet, then Souporcell, then scSplit.
  -p PCT_AGREEMENT, --pct_agreement PCT_AGREEMENT
                        The proportion of a cluster that match the 'ref'
                        assignment to assign that cluster the individual
                        assignment from the reference. Can be between 0.5 and
                        1. Default is 0.9.
  -m METHOD, --method METHOD
                        Combination method. Options are 'MajoritySinglet'.
                        'AtLeastHalfSinglet', 'AnySinglet' or 'AnyDoublet'. We
                        have found that 'MajoritySinglet' provides the most
                        accurate results in most situations and therefore
                        recommend this method. See https://demultiplexing-
                        doublet-detecting-
                        docs.readthedocs.io/en/latest/CombineResults.html for
                        detailed explanation of each intersectional method.
                        Leave blank if you just want all the softwares to be
                        merged into a single dataframe.

Combination Methods - Additional Information

There are four options for making combined droplet type (singlet or doublet) and individual assignment from the softwares used:

• MajoritySinglet

  • If more than half of the softwares identify a droplet as a singlet, it is classified as a singlet.

  • If more than half the demultiplexing softwares identify the same indivdual, that assignment is used for the droplet.

  • We have found

• AtLeastHalfSinglet

  • If at least half of the softwares identify a droplet as a singlet, it is classified as a singlet.

  • If at least half the demultiplexing softwares identify the same indivdual, that assignment is used for the droplet.

• AnySinglet

  • If this droplet is identified as a singlet by any software, the droplet is classified as a singlet.

  • In other words, a doublet is only called if all softwares identified that droplet as a doublet.

• AnyDoublet

  • A droplet is classified as a singlet only if all softwares identify it as a singlet.

  • In other words, a doublet is called if any software identifies that droplet as a doublet.

An example command that combines Demuxlet results, Souporcell results, Solo results and Scds results would look like this: There are a two different options for using this script:

Combine Results + Joint Droplet CallsCombine Results

The first option is to select a method to make joint calls on the individual assignment and singlet-doublet droplet types using the softwares included.

singularity exec Demuxafy.sif Combine_Results.R \
  -o $OUTDIR/combined_results.tsv \
  --demuxlet $DEMUXLET_OUTDIR \
  --souporcell $SOUPORCELL_OUTDIR \
  --solo $SOLO_OUTDIR \
  --scds $SCDS_OUTDIR \
  --method "MajoritySinglet"

Note

The path to the directories will work if the file names are the expected file names based on the example tutorials. However, if you used a different file naming convention or changed the names, you can also provide the full path to the exact file for each software.

Results and Interpretation

After running the Combine_Results.R script, you should have two, three or four files depending on if you used demultiplexing softwares and if you used joint droplet calling. Here, we show the results for the above example that also provides combined calls with the “MajoritySinglet” calls.

.
├── combined_results_demultiplexing_summary.tsv
├── combined_resultsSinglets_upset.pdf
├── combined_results_summary.tsv
├── combined_results.tsv
└── combined_results_w_combined_assignments.tsv

Note

• You will only have the combined_results_demultiplexing_summary.tsv file if you included demultiplexing softwares.

• And you will only have the combined_results_w_combined_assignments.tsv file if you ran it with --method

Here’s a deeper look at the contents of each of these results:

• combined_resultsSinglets_upset.pdf

  • This is an upset figure of the droplets which are colored by their finall individual or doublet classification.

  • A filled circle indicates the that those droplets are classified as singlets by that method while empty circles indicate a doublet classification by that software

  

• combined_results.tsv

  • Has the selected results combined; only including key columns.

    Barcode	Demuxlet_DropletType	Demuxlet_Individual_Assignment	Souporcell_Individual_Assignment	Souporcell_Cluster	Souporcell_DropletType	scds_score	scds_DropletType	solo_DropletType	solo_DropletScore
    AAACCTGAGATAGCAT-1	singlet	41_41	41_41	6	singlet	0.116344358493288	singlet	singlet	-8.442187
    AAACCTGAGCAGCGTA-1	singlet	465_466	465_466	11	singlet	0.539856378453988	singlet	singlet	-2.8096201
    AAACCTGAGCGATGAC-1	singlet	113_113	113_113	5	singlet	0.0237184380134577	singlet	singlet	-2.8949203
    AAACCTGAGCGTAGTG-1	singlet	349_350	349_350	3	singlet	0.163695865366576	singlet	singlet	-5.928284
    AAACCTGAGGAGTTTA-1	singlet	632_633	632_633	7	singlet	0.11591462421927	singlet	doublet	0.2749935
    AAACCTGAGGCTCATT-1	singlet	39_39	39_39	12	singlet	0.0479944175570073	singlet	singlet	-5.2726507
    AAACCTGAGGGCACTA-1	singlet	465_466	465_466	11	singlet	0.374426050641161	singlet	singlet	-0.65760195
    AAACCTGAGTAATCCC-1	singlet	660_661	660_661	4	singlet	0.247842972104563	singlet	singlet	-3.5948637
    AAACCTGAGTAGCCGA-1	doublet	doublet	unassigned	unassigned	unassigned	0.342998285281922	singlet	singlet	-0.50507957
    …	…	…	…	…	…	…	…	…	…

• combined_results_summary.tsv

  • The number of each of the combinations of the software cell type classifications

  Demuxlet_DropletType	Souporcell_DropletType	scds_DropletType	solo_DropletType	N
  singlet	singlet	singlet	singlet	16193
  doublet	doublet	doublet	doublet	1714
  singlet	singlet	singlet	doublet	947
  doublet	doublet	singlet	singlet	468
  singlet	singlet	doublet	singlet	392
  singlet	singlet	doublet	doublet	345
  doublet	doublet	singlet	doublet	335
  doublet	singlet	singlet	singlet	171
  doublet	doublet	doublet	singlet	169
  doublet	singlet	doublet	doublet	114
  doublet	singlet	singlet	doublet	44
  doublet	singlet	doublet	singlet	18
  singlet	doublet	singlet	singlet	17
  singlet	unassigned	singlet	singlet	13
  doublet	unassigned	singlet	singlet	11
  singlet	doublet	doublet	doublet	9
  singlet	doublet	singlet	doublet	6
  singlet	doublet	doublet	singlet	5
  doublet	unassigned	singlet	doublet	4
  doublet	unassigned	doublet	doublet	3
  doublet	unassigned	doublet	singlet	2
  unassigned	unassigned	singlet	singlet	2

• combined_results_demultiplexing_summary.tsv

  • Summary of the number of each of the combination of classifications by demultiplexing software:

    Demuxlet_Individual_Assignment	Souporcell_Individual_Assignment	N
    doublet	doublet	2706
    352_353	352_353	1603
    43_43	43_43	1547
    597_598	597_598	1510
    349_350	349_350	1450
    42_42	42_42	1417
    660_661	660_661	1358
    113_113	113_113	1333
    39_39	39_39	1289
    …	…	…

• combined_results_w_combined_assignments.tsv

  • Dataframe combining all the software results together + combined assignment based on selected method:

  Barcode	Demuxlet_DropletType	Demuxlet_Individual_Assignment	Souporcell_Cluster	Souporcell_Individual_Assignment	Souporcell_DropletType	scds_score	scds_DropletType	solo_DropletType	solo_DropletScore	MajoritySinglet_DropletType	MajoritySinglet_Individual_Assignment
  AAACCTGAGATAGCAT-1	singlet	41_41	6	41_41	singlet	0.116344358493288	singlet	singlet	-8.442187	singlet	41_41
  AAACCTGAGCAGCGTA-1	singlet	465_466	11	465_466	singlet	0.539856378453988	singlet	singlet	-2.8096201	singlet	465_466
  AAACCTGAGCGATGAC-1	singlet	113_113	5	113_113	singlet	0.0237184380134577	singlet	singlet	-2.8949203	singlet	113_113
  AAACCTGAGCGTAGTG-1	singlet	349_350	3	349_350	singlet	0.163695865366576	singlet	singlet	-5.928284	singlet	349_350
  AAACCTGAGGAGTTTA-1	singlet	632_633	7	632_633	singlet	0.11591462421927	singlet	doublet	0.2749935	singlet	632_633
  AAACCTGAGGCTCATT-1	singlet	39_39	12	39_39	singlet	0.0479944175570073	singlet	singlet	-5.2726507	singlet	39_39
  AAACCTGAGGGCACTA-1	singlet	465_466	11	465_466	singlet	0.374426050641161	singlet	singlet	-0.65760195	singlet	465_466
  AAACCTGAGTAATCCC-1	singlet	660_661	4	660_661	singlet	0.247842972104563	singlet	singlet	-3.5948637	singlet	660_661
  AAACCTGAGTAGCCGA-1	doublet	doublet	unassigned	doublet	doublet	0.342998285281922	singlet	singlet	-0.50507957	doublet	doublet
  …	…	…	…	…	…	…	…	…	…	…	…

Citation

If you used the Demuxafy platform for analysis, please reference our paper (REFERENCE).