2. cell_calling.py

2.1. Description

Call cells from background.

• First, calculates the “read count” and “UMI count” for each barcode (i.e. cell barcode),and generates the barcode rank plot and density plot.
• Then, using the Bayesian Gaussian Mixture Model (BGMM) to classify barcodes into “cell-associated” and “background-associated”.

Options:

--version	show program’s version number and exit
-h, --help	show this help message and exit
-i IN_FILE, --infile=IN_FILE
	Input file in BAM foramt.
-o OUT_FILE, --outfile=OUT_FILE
	The prefix of output files.
--cb-tag=CB_TAG
	Tag of error-corrected cellular barcode in BAM file. default=’CB’
--umi-tag=UMI_TAG
	Tag of error-corrected UMI in BAM file. default=’UB’
--cb-num=CB_LIMIT
	Maximum cell barcodes (ranked by associated UMI frequency) analysed. default=100000
--min-read-count=MIN_READS
	The minimum number of reads to filter out cell barcode. default=200
-r, --report	If set, generates report file for mixture models. default=False
-s RANDOM_STATE, --seed=RANDOM_STATE
	The seed used by the random number generator. default=0
--prob-cut=PROBABILITY_CUTOFF
	The probabiilty cutoff [0.5, 1] to assign cell barcode to the “cell” or the “background” component. default=0.5
--verbose	If set, print detailed information for debugging.

2.2. Example

$ python3 cell_barcode.py -i possorted_genome_confident.bam -o output

2020-10-05 02:44:42 [INFO]  Top 100000 cell barcodes (ranked by associated UMI frequency) will be analyzed.
2020-10-05 02:44:42 [INFO]  Only count UMIs for cell barcodes with more than 200 reads.
2020-10-05 02:44:42 [INFO]  Reading BAM file "possorted_genome_confident.bam". Count reads for each cell barcode ...
2020-10-05 02:59:37 [INFO]  Total 98839578 alignments processed
2020-10-05 02:59:37 [INFO]  Filtering cell barcodes ...
2020-10-05 02:59:37 [INFO]  Total cell barcode: 856517
2020-10-05 02:59:37 [INFO]  Cell barcode with more than 200 reads: 8204
2020-10-05 02:59:37 [INFO]  Reading BAM file "possorted_genome_confident.bam". Count UMIs for each cell barcode ...
2020-10-05 03:15:33 [INFO]  Total 90458232 alignments processed
2020-10-05 03:15:33 [INFO]  Writing cell barcodes' reads and UMI frequencies to "output.Read_UMI_freq.tsv"
2020-10-05 03:15:33 [INFO]  Done.
2020-10-05 03:15:38 [INFO]  Read output.Read_UMI_freq.tsv to build Bayesian Gaussian Mixture Model (BGMM)...
2020-10-05 03:15:38 [INFO]  Reading input file: "output.Read_UMI_freq.tsv"
2020-10-05 03:15:38 [INFO]     Total analyzed barcodes: 8204
2020-10-05 03:15:38 [INFO]  Build BGMM ...
2020-10-05 03:15:38 [INFO]  Building Bayesian Gaussian Mixture model for subject: UMI_count ...
2020-10-05 03:15:40 [INFO]  Building Bayesian Gaussian Mixture model for subject: read_count ...
2020-10-05 03:15:40 [INFO]  Classify cell barcode using the BGMM models ...
2020-10-05 03:15:40 [INFO]  Writing to "output.UMI_count_classification.txt" ...
2020-10-05 03:15:40 [INFO]  Writing to "output.read_count_classification.txt" ...
2020-10-05 03:15:40 [INFO]  Writing R script to "output.Read_UMI_freq.r"

2.3. Output files

1. output.read_count_classification.txt and output.UMI_count_classification.txt. Classify cell barcodes according to read (or UMI) count.

• Column-1 : The barcode sequence.
• Column-2 : Read/UMI count in log10 scale.
• Column-3 : The probability that this barcode belonging to “background” group.
• Column-4 : The probability that this barcode belonging to “cell” group.
• Column-5 : Assigned lable (“cell”, “unknown” or “background”).

Barcode	log10_count	background_prob	cell_prob	Lable
AACAACCCAGTTCTAG	5.68488933	1.03E-73	1	cell
CTATCCGCATGGATCT	5.61115166	2.05E-70	1	cell
AATCGTGTCTTTGATC	5.55137314	8.49E-68	1	cell
…
GAACGTTTCGGCAGTC	2.30103	0.98811857	0.01188143	background
CATCCACAGGCGAAGG	2.32633586	0.98862923	0.01137077	background
TCACGGGGTGATATAG	2.31175386	0.98836594	0.01163406	background

2. output.Read_UMI_freq.tsv

Serial#	Cell_barcode	read_count	UMI_count
1	AACAACCCAGTTCTAG	484049	202124
2	CTATCCGCATGGATCT	408462	184155
3	AATCGTGTCTTTGATC	355937	165832
4	CTTCTCTGTTGTCCCT	349352	159697
5	GAGAAATCATGACACT	362654	153856
…

3. output.Read_UMI_freq.r

Generate figure as below. Panels A, B, C were generated from read count, and panels D, E, F were generated from UMI count. (A) and (D) Density plots of cell barcodes that have been classified into background (red) and cells (blue). (B) and (E) Barcode rank plots of all cell barcodes. (C) and (F) Probability rank plots of all cell barcodes.

Note

All the cell barcodes were classified into two groups, because the default --prob-cut is 0.5. In this Scenario, each barcode was either classified into cell-associated group (when prob >= 0.5) or background-associated group (when prob < 0.5). If the --prob-cut is set to 0.95, cell barcodes will be classified into three groups: “cell-associated” (when cell_prob >= 0.95), “background-associated” (when background_prob >= 0.95) and “unknown” (cell_prob < 0.9 and background_prob < 0.95). The “unknown” group might contains cells with low RNA content. Please see figure below.