This is a small library and a bunch of clients to perform various operations on FASTQ files (such as merging partial overlaps, or quality filtering). For now it only works with,

• Illumina output files generated by CASAVA version 1.8.0 or higher (it will support earlier versions of CASAVA soon),
• Paired-end runs.

In this README file you will find information about the following items:

• Obtaining the source code.
• Config file format library requires.
• Merging partially overlapping illumina pairs.
• Quality filtering script for "Complete Overlap" approach described by Eren et al.
• Quality filtering script that uses the method described by Minoche et al.
• Quality filtering script that uses the method described by Bokulich et al.

You can get in touch with me via meren at mbl dot edu.

• Contents
• Obtaining the Source Code
  • Requirements
• Config File Format
  • [general] section
  • [files] section
  • [prefixes] section
• Merging Partially Overlapping Illumina Pairs
  • Example STATS output
  • Recovering high-quality reads from merged reads file
• Quality Filtering
  • "Complete Overlap" analysis for V6
    • Example STATS output
  • Minoche et al.
    • Example STATS output
    • Example PNG files
  • Bokulich et al.
    • Example STATS output:
    • Example PNG files
• Questions?

# Obtaining the Source Code

You can create a copy of the codebase by simply installing git and running this command in your terminal window:

 git clone git://github.com/meren/illumina-utils.git

This will generate the illumina-utils directory within the directory from which you run this command. Or you can simply download the zipped codebase through your browser and unzip it. Although it is not mandatory, having git installed and learning how to use it have advantages such as being able to keep your copy updated by synchronizing it with the master repository by simply typing git pull in your terminal window. However, if you do not wish to use git, you can always reach the zipped archive file of the latest version of the codebase via this link:

 https://github.com/meren/illumina-utils/archive/master.zip

Once you have the codebase, you should update your PYTHONPATH and PATH environment variables for easy access to the scripts and required libraries. You can do it by adding the following two lines in your .bash_profile file (it is a hidden file in your home directory, you can reach it by opening a terminal window and typing "nano ~/.bash_profile"):

 export PYTHONPATH=$PYTHONPATH:/path-to/illumina-utils/
 export PATH=$PATH:/path-to/illumina-utils/scripts

Please remember to replace path-to place holder in the lines above with the actual path that points the where illumina-utils directory is on your file system.

## Requirements

In order to use this software package fully, you need following items available on your system:

• matplotlib (required for visualizations)
• python-Levenshtein (required for fast merge of partially overlapping reads)
• EMBOSS (merger is a part of EMBOSS software distribution and it is used for alignment of partially overlapping reads (when --slow-merge option is not used))
• R (required for visualizations today and will be used for statistical analyses)
  • ggplot2 (the R module that needs to be installed for R requirement)

Clients under the scripts directory require information to be passed along with the same config file format, if they require a config file as an input. Following is a config file template (there is also a sample file in the codebase):

[general]
project_name = project name
researcher_email = [email protected]
input_directory = /full/path/test_input
output_directory = /full/path/test_output


[files]
pair_1 = pair_1_aaa, pair_1_aab, pair_1_aac, pair_1_aad, pair_1_aae, pair_1_aaf 
pair_2 = pair_2_aaa, pair_2_aab, pair_2_aac, pair_2_aad, pair_2_aae, pair_2_aaf

[prefixes]
pair_1_prefix = ^....TACGCCCAGCAGC[C,T]GCGGTAA.
pair_2_prefix = ^CCGTC[A,T]ATT[C,T].TTT[G,A]A.T

## [general] section

This is a mandatory section that contains project_name, researcher_email, input_directory and output_directory directives.

Two critical declerations in [general] section are input_directory and output_directory:

• input_directory: Full path to the directory where FASTQ files reside.
• output_directory: Full path to the directory where the output of the operation you will perform on this config to be stored. Since when it is Illumina we are dealing with huge files, the codebase is pretty conservative to protect users from making simple mistakes which may result in huge losses. So, if you don't create the output_directory, you will get an error (it will not be automatically generated). If there is already a file in the output_directory with the same name with one of the outputs, you will get an error (it will not be overwritten). project_name will be used as a prefix for the naming convention of output files, so it would be wise to choose something descriptive and UNIX-compatible.

## [files] section

files section is where you list your file names to be found in the output_directory. Each file name has to be comma separated. The index of each file name in the comma seperated list, must match with its pair in the second list (see the example config file above).

prefixes section is optional. If you have barcodes and primers in your reads, and you want them to be trimmed, you can use regular expressions to specify them. If prefixes are defined, results would contain only pairs that matched them.

Pairs generated by paritally overlapping library preperation can be merged using merge-illumina-pairs. Once you create your config file, you simply call it with the config file as a parameter.

By default, merging program uses Levenshtein distance to find the best merging strategy for two reads in a pair, starting from the minimum expected overlap (15 nt is the default, and can be changed through the appropriate command line parameter).

Default merging strategy requires Levenshtein module to be installed. Other merging alternative is using Needleman-Wunsch algorithm for the alignment of two reads in a pair. It can be done by declaring --slow-merge as a parameter. However it is, expectedly, very slow. Needleman-Wunsch alignment performs better only if there are a lot of insertion/deletion errors present in the dataset, which is very uncommon in Illumina HiSeq and MiSeq results.

merge-illumina-pairs will create FASTA files for reads that were merged successfuly, or failed to merge. In the FASTA file for merged reads, the length of the overlapped region and the number of mismatches found in the overlapped part will be reported in the header line for each entry. The place of mismatch will be shown with capital letters in the sequences.

An example header line from the FASTA file for merged reads is shown below:

>M01028:4:000000000-A1Y0P:1:1101:18829:1947 1:N:0:1|o:83|m/o:0.048193|MR:n=0;r1=2;r2=2|Q30:p,77;p,76|mismatches:4

Each field is separated from each other by a "|" character. Field one is the original defline from the FASTQ file of read 1. Following items explain details of these fields and command line options that affect them:

• o: Length of the overlap.
• m/o: The P value. P value is the ratio of the number of mismatches and the length of the overlap. Merged sequences can be discarded based on this ratio. The default is 0.3. This value should be changed through the command line parameter --p-value depending on the expected overlap size (if the expected length of overlap is 100 nts and if you choose to eliminate any pairs with more than 5 mismatches at the overlapping region, you can set the --p-value parameter to 0.05).
• MR: "Mismatches Recovered". When there is a mismatch in the overlapped region, the base to be used in the final merged sequence is picked from whichever read possesses the higher Q-score (and shown as a capital letter in the merged sequence). If a mismatch is recovered from read 1, it increases the number next to r1 in this field, and so forth. However, if there is a disagreement between two reads, and neither of the reads have a Q-score higher than a minimum acceptable value, the corresponding base denoted with an N in the merged read, and the number next to none is increased by one. By default, the minimum Q-score value is 10. This value can be changed via the command line parameter --min-qual-score. Note that if --ignore-Ns flag is not declared, all merged sequences that had at least one disagreement which can't be recovered from neither reads due to --min-qual-score value will be discarded.
• Q30: By default, quality filtering is being done based only on the mismatches found in the overlapped region, and the beginning and the end of merged reads are not being checked. However a final control can be enforced using the command line flag --enforce-Q30-check. This flag turns on the Q30 check, as it was explained by Minoche et al. Briefly, Q30-check eliminates pairs if the 66% of bases in the first half of each read do not have Q-scores over Q30. Note that Q30 is applied only to the parts of reads that did not overlap. If either of reads fail Q30 check, merged sequence is discarded. p,77;p,76 in the example header reads as "read 1 passed Q30 check (threfore p, failed case denoted by an f), and 77 bases in the first half of it had a better Q-score than 30; read 2 passed Q30 check, and 76 bases in the first half of it had a better Q-score than 30". If Q30-check was not enforced n/a appears next to it.
• mismatches: Number of mismatches at the overlapped region for quick filtering of resulting reads.

Here is a snippet from the merged sequences file (reads are trimmed from both ends for readability):

>M01028:4:000000000-A1Y0P:1:1101:15704:1943 1:N:0:1|o:87|m/o:0.022989|MR:n=0;r1=2;r2=0|Q30:p,77;p,72|mismatches:2
[...]ggtagatggaatataacatgtagcggtgaaatGctTagatatgttatggaacaccgattgcgaaggcagtctactaagtcgatattgacgctgaggcacgaaagcgtgggtagcgaacag[...]
>M01028:4:000000000-A1Y0P:1:1101:18231:1947 1:N:0:1|o:86|m/o:0.058140|MR:n=0;r1=5;r2=0|Q30:p,74;p,66|mismatches:5
[...]ggaaagtggaatttctaGTGTagaggtgaaattcgtagatattagaaagaacatcaaaggcGaaggcaactttctggatcattactgacactgaggaacgaaagcatgggtagcgaagag[...]
>M01028:4:000000000-A1Y0P:1:1101:18829:1947 1:N:0:1|o:83|m/o:0.048193|MR:n=0;r1=2;r2=2|Q30:p,77;p,76|mismatches:4
[...]ggggggtagaatTccacgtgtagcagtgaaatgcgtagagatgtggaGgaatAtcaatggcgaaggcagccccctgggataacactgacgCtcatgcacgaaagcgtggggagcgaacag[...]

If the program runs successfully, these files will appear in the output_directory:

• project_name_MERGED (successfuly merged reads)
• project_name_FAILED (failed sequences due to m/o)
• project_name_FAILED_WITH_Ns (failed merged sequences for having ambiguous bases)
• project_name_FAILED_Q30 (failed merged sequences for not passing Q30-check, if enforced)
• project_name_MISMATCHES_BREAKDOWN (number of mismatches breakdown)
• project_name_STATS (numbers regarding the run)

project_name_MISMATCHES_BREAKDOWN file can be visualized using the R script, mismatches-distribution.R, included in the codebase (it will require ggplot2 to be available on the system). Here is an example:

When merge-illumina-pairs is run with --compute-qual-dicts it will also generate visualization of quality scores for different number of mismatch levels. Please see command line options for more information.

The project_name_STATS file that is created in the output directory contains important information about the merging operation. It is a good practice to check the numbers and make sure there is no anomalies. Here is an example output:

Number of pairs analyzed        	2500
Prefix failed in read 1         	0
Prefix failed in read 2         	0
Prefix failed in both           	0
Passed prefix total             	2500
Failed prefix total             	0
Merged total                    	1479
Merge failed total              	1021
Merge discarded due to P        	598
Merge discarded due to Ns       	348
Merge discarded due to Q30      	75
Total number of mismatches      	13101
Mismatches recovered from read 1	10360
Mismatches recovered from read 2	1413
Mismatches replaced with N      	1328

Mismatches breakdown:

0	372
1	326
2	225
3	154
4	120
5	86
6	70
7	49
8	40
9	21
10	11
11	4
12	1


Command line            	merge-illumina-pairs miseq_partial_overlap_config.ini z --enforce --compute
Work directory              	/path/to/the/working/directory
"p" value                   	0.300000
Min overlap size            	15
Min Q-score for mismatches  	10
Ns ignored?                 	False
Q30 enforced?               	True
Slow merge?                 	False

## Recovering high-quality reads from merged reads file

If merge-illumina-pairs finishes successfuly, it will generate project_name_MERGED for successfuly merged reads. A successful merge depends on the o/r value, Q30-check and lack of ambiguous bases in the merged sequence. However, succesfully merged reads based on user-defined or default parameters may not be as accurate as needed depending on the project. Further elimination of reads can be done by filtering out reads based on the number of mismatches they present at the overlapped region. For instance, user can decide to use only merged sequences with 0 mismatches from the resulting FASTA file.

Program filter-merged-reads can be used to retain high-quality reads from project_name_MERGED file. To retain reads with 0 mismatches at the overlapped region you can simply run this command on your project_name_MERGED to generate a file with filtered reads project_name_FILTERED:

 filter-merged-reads project_name_MERGED --max-mismatches 0 --output project_name_FILTERED

Resulting file would be the file to use in downstream analyses.

## "Complete Overlap" analysis for V6

analyze-illumina-v6-overlaps can be used to generate very high quality short reads from sequences that were generated by a short insert size library preperation method. Library preperation method and the efficacy of the complete overlap analysis is described in Eren et al. Once the manuscript is published, the reference will be available here. The output of the analyze-illumina-v6-overlaps script include these files:

• project_name-STATS.txt (an example output can be seen below)
• project_name-PERFECT_reads.fa (FASTA file for reads that passed the complete overlap analysis)
• project_name-Q_DICT.cPickle.z (gzipped cPickle object for Python that holds the machine reported quality scores for each group of failed and passed reads)
• project_name-READ_IDs.cPickle.z (gzipped cPickle object for Python that holds the read fate information)

If the program is run with --visualize-quality-curves option, following files will also be generated in the output directory:

• project_name-PASSED.png (visualization of mean machine reported quality scores per tile for pairs that passed the the complete overlap analysis)
• project_name-FAILED_MISMATCH.png (same for pairs that failed due to one or more mismatches at the region of read of interest)
• project_name-FAILED_RP.png (same for pairs that lacked a proper reverse primer)
• project_name-FAILED_FP.png (same for pairs that lacked a proper forward primer)

$ cat 9022_B9-STATS.txt
number of pairs                 : 828243
total pairs passed              : 618589 (%74.69 of all pairs)
  perfect pairs with Ns         : 0 (%0.00 of perfect pairs)
  recovered ambiguous bases (p1): 0 (%0.00 of perfect pairs)
  recovered ambiguous bases (p2): 0 (%0.00 of perfect pairs)
total pairs failed              : 209654 (%25.31 of all pairs)
  FP failed in both pairs       : 5086 (%2.43 of all failed pairs)
  FP failed only in pair 1      : 386 (%0.18 of all failed pairs)
  FP failed only in pair 2      : 116822 (%55.72 of all failed pairs)
  RP failed in both pairs       : 7076 (%3.38 of all failed pairs)
  RP failed only in pair 1      : 6767 (%3.23 of all failed pairs)
  RP failed only in pair 2      : 7647 (%3.65 of all failed pairs)
  FAILED_MISMATCH               : 65870 (%31.42 of all failed pairs)
  FAILED_RP                     : 21490 (%10.25 of all failed pairs)
  FAILED_FP                     : 122294 (%58.33 of all failed pairs)

Quality filtering suggestions made by Minoche et al is implemented in analyze-illumina-quality-minoche script. The output of the scripts include these files:

• project_name-STATS.txt (file that contains all the numbers about quality filtering process, an example output can be seen below)
• project_name-QUALITY_PASSED_R1.fa (pair 1's that passed quality filtering)
• project_name-QUALITY_PASSED_R2.fa (matching pair 2's)
• project_name-READ_IDs.cPickle.z (gzipped cPickle object for Python that keeps the fate of read IDs, this file may be required by other scripts in the library for purposes such as visualization, or extracting a particular group of reads from the original FASTQ files)

If the program is run with --visualize-quality-curves option, these files will also be generated in the output directory:

• project_name-PASSED.png (visualization of mean quality scores per tile for pairs that passed the quality filtering)
• project_name-FAILED_REASON_C33.png (visualization of mean quality scores per tile for pairs that failed quality filtering due to C33 filtering (C33: less than 2/3 of bases were Q30 or higher in the first half of the read following the B-tail trimming))
• project_name-FAILED_REASON_N.png (same above, but for pairs that contained an ambiguous base after B-tail trimming)
• project_name-FAILED_REASON_P.png (same above, but for pairs that were too short after B-tail trimming)
• project_name-Q_DICT.cPickle.z (gzipped cPickle object for Python that holds mean quality scores for each group of reads)

$ cat 9022_B9-STATS.txt
number of pairs analyzed      : 122929
total pairs passed            : 109041 (%88.70 of all pairs)
  total pair_1 trimmed        : 6476 (%5.94 of all passed pairs)
  total pair_2 trimmed        : 9059 (%8.31 of all passed pairs)
total pairs failed            : 13888 (%11.30 of all pairs)
  pairs failed due to pair_1  : 815 (%5.87 of all failed pairs)
  pairs failed due to pair_2  : 12193 (%87.80 of all failed pairs)
  pairs failed due to both    : 880 (%6.34 of all failed pairs)
  FAILED_REASON_P             : 12223 (%88.01 of all failed pairs)
  FAILED_REASON_N             : 38 (%0.27 of all failed pairs)
  FAILED_REASON_C33           : 1627 (%11.72 of all failed pairs)

Quality filtering suggestions made by Bokulich et al is implemented in analyze-illumina-quality-bokulich script. The output of the scripts include these files:

• project_name-STATS.txt
• project_name-QUALITY_PASSED_R1.fa
• project_name-QUALITY_PASSED_R2.fa
• project_name-READ_IDs.cPickle.z

If the program is run with --visualize-quality-curves option, these files will also be generated in the output directory:

• project_name-PASSED.png
• project_name-FAILED_REASON_P.png (visualization of mean quality scores per tile for pairs that failed quality filtering for being too short after quality trimming)
• project_name-FAILED_REASON_N.png (same above, but having more ambiguous bases than n after quality trimming)
• project_name-Q_DICT.cPickle.z

number of pairs analyzed      : 122929
total pairs passed            : 111598 (%90.78 of all pairs)
  total pair_1 trimmed        : 1994 (%1.79 of all passed pairs)
  total pair_2 trimmed        : 9227 (%8.27 of all passed pairs)
total pairs failed            : 11331 (%9.22 of all pairs)
  pairs failed due to pair_1  : 738 (%6.51 of all failed pairs)
  pairs failed due to pair_2  : 10159 (%89.66 of all failed pairs)
  pairs failed due to both    : 434 (%3.83 of all failed pairs)
  FAILED_REASON_P             : 11299 (%99.72 of all failed pairs)
  FAILED_REASON_N             : 32 (%0.28 of all failed pairs)

Please don't hesitate to get in touch with me via meren at mbl dot edu.