# Downsample data by barcode being deprecated

DEPRECATION NOTICE

The downsample module is being replaced with djinn, which is provided with a conda-based Harpy installation as of version 3.1 (build 2). It will be officially removed from Harpy starting with version 3.3.

One of either:
- one alignment file .bam .sam case insensitive
- one set of paired-end reads in FASTQ format .fq .fastq ❤️ gzipped recommended case insensitive
Barcodes in a SAM tag (e.g. BX:Z:) for both BAM and FASTQ inputs
- See Section 1 of the SAM Spec here for details

While downsampling (subsampling) FASTQ and BAM files is relatively simple with tools such as awk, samtools, seqtk, seqkit, etc., downsample allows you to downsample a BAM file (or paired-end FASTQ) by barcodes. That means you can keep all the reads associated with d number of barcodes or -d fraction of barcodes (e.g. -d 0.5 will downsample to 50% of all barcodes).

usage
# BAM file
harpy downsample OPTIONS... BAM

# FASTQ files
harpy downsample OPTIONS... FASTQ1 FASTQ2

example | downsample bam file, allowing only 30% invalid barcodes to be sampled
harpy downsample -d 1000 -b BC -i 0.3 -p sample1.sub1000 sample1.bam

example | downsample fastq pair, skipping invalid barcodes
harpy downsample -d 1000 -i 0 -p sample1.sub1000 sample1.F.fq.gz sample1.R.fq.gz

# Running Options

In addition to the common runtime options , the downsample module is configured using the command-line arguments below.

argument	default	description
`INPUT`		required One BAM file or both read files from a paired-end FASTQ pair
`--barcode-tag` `-b`	`BX`	SAM tag that contains the barcode
`--downsample` `-d`		required Number/fraction of barcodes to downsample to
`--invalid` `-i`	`1`	Proportion of invalid barcodes to potentially sample
`--prefix` `-p`	`downsampled`	Prefix for output files
`--random-seed`		Random seed for sampling optional

# invalid barcodes

The --invalid options determines what proportion of invalid barcodes appear in the barcode pool that gets subsampled. The --invalid proportion doesn't necessarily reflect how many end up getting sampled, rather what proportion will be considered for sampling. The proportions equate to:

0: invalid barcodes are skipped altogether
1: all invalid barcodes appear in the barcode pool that gets subsampled
0<i<1: that proportion of barcodes appear in the barcode pool that gets subsampled

# Downsample Workflow

graph LR
    subgraph fastq
        R1([read 1]):::clean---R2([read 2]):::clean
    end
    subgraph bam
        bamfile([bam]):::clean
    end
    fastq-->|bam conversion|bam
    bam-->sub([extract and\n subsample barcodes]):::clean
    sub-->exreads([extract reads]):::clean
    bam-->exreads
    fastq-->exreads
    style fastq fill:#f0f0f0,stroke:#e8e8e8,stroke-width:2px,rx:10px,ry:10px
    style bam fill:#f0f0f0,stroke:#e8e8e8,stroke-width:2px,rx:10px,ry:10px
    classDef clean fill:#f5f6f9,stroke:#b7c9ef,stroke-width:2px

linked-read