Skip to content

Commit

Permalink
1.0.0 update
Browse files Browse the repository at this point in the history
  • Loading branch information
zkronenberg committed Apr 23, 2021
1 parent 3f3f661 commit 204078b
Show file tree
Hide file tree
Showing 2 changed files with 314 additions and 70 deletions.
136 changes: 66 additions & 70 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,19 +8,18 @@

PacBio Amplicon Analysis (_pbaa_) separates complex mixtures of amplicon targets from genomic samples. The _pbaa_ application is designed to cluster and generate high-quality consensus sequences from HiFi reads. This application only works on HiFi amplicon data. There are several assumptions made within the code that will only support high quality reads (>QV20). This application will not work on CLR data. _pbaa_ is reference aided method (pseudo de-novo).

Typical use cases involve multi-allelic samples where the sample-specific ploidy or copy number is unknown. _pbaa_ can effectively separate alleles with one to many variants, including SNVs and large indels contained within the target region. _pbaa_ has been optimized and tested for datasets with a moderate (<10) cluster count. Feedback for higher cluster density is welcome and may be addressed in future releases.
Typical use cases involve multi-allelic samples where the sample-specific ploidy or copy number is unknown. _pbaa_ can effectively separate alleles with one to many variants, including SNVs and large indels contained within the target region. _pbaa_ has been optimized and tested for datasets with a moderate to high (<50) cluster counts.

## Workflow
![HiFi Amplicon Analysis Workflow](img/workflow.png)
![HiFi Amplicon Analysis Workflow](img/v1.0.0.png)

## Availability
Latest version can be installed via bioconda package `pbaa`.
The latest version can be installed via bioconda package `pbaa`.

Please refer to our [official pbbioconda page](https://github.com/PacificBiosciences/pbbioconda)
for information on Installation, Support, License, Copyright, and Disclaimer.

## Latest Version
Version **0.1.2**: [Full changelog here](#full-changelog)
[Full changelog here](#full-changelog)

## Usage
_pbaa_ has two executables, cluster and bam paint.
Expand Down Expand Up @@ -49,38 +48,46 @@ pbaa cluster - Run clustering tool.
Usage:
pbaa cluster [options] <guide input> <read input> <prefix>
guide input FILE Guide sequence(s) in fasta format. A FOFN can be provide for multiple files.
read input FILE De-multiplexed HiFi reads in fastq format. A FOFN can be provide for multiple files.
prefix STR Output prefix for run.
guide input FILE Guide sequence(s) in fasta format indexed with samtools faidx version 1.9 or
greater. A FOFN can be provide for multiple files.
read input FILE De-multiplexed HiFi reads in fastq format indexed with samtools fqidx version 1.9
or greater. A FOFN can be provide for multiple files.
prefix STR Output prefix for run.
Placement and Variant Options:
--filter INT Low coverage kmer count cutoff. Automatically adjusted by min-var-frequency. [3]
--trim-ends INT Number of bases to trim from both sides of reads during graph construction. [20]
--min-var-frequency FLOAT Minimum variant frequency. [0.05]
--filter INT Variants with coverage lower than filter will be ignored. [3]
--trim-ends INT Number of bases to trim from both sides of reads during graph construction and
variant detection. [5]
--pile-size INT The number of best alignments to keep for each read during error correction. [30]
--min-var-frequency FLOAT Minimum coverage frequency within a pile. [0.3]
--max-alignments-per-read INT The number of random alignments, for each read, within a guide grouping [1000]
Clustering Options:
--fixed-cluster-count INT Maximum number of clusters per locus/guide-group. [10]
--em-iterations INT Number of iterations to run expectation maximization. [300]
--no-cluster-by-length Disable fallback length clustering if no variants were discovered.
--max-reads-per-guide INT The number randomly selected reads to use within a guide grouping. [500]
--iterations INT Number of iterations to run k-means. [9]
--seed INT Randomization seed. [1984]
Consensus Options:
--min-cluster-frequency FLOAT Low frequency cluster cutoff. [0.1]
--min-cluster-read-count INT Low read count cluster cutoff. [5]
--max-consensus-reads INT Maximum number of reads to use per cluster consensus. [100]
--off-target-groups STR Group names to exclude, i.e. these loci are off-target (not amplified).
--max-consensus-reads INT Maximum number of reads to use per cluster consensus. [100]
Filtering Options:
--max-amplicon-size INT Upper read length cutoff, longer reads will be skipped. [15000]
--min-read-qv FLOAT Low read QV cutoff. [20]
--off-target-groups STR Group names to exclude, i.e. these loci are off-target (not amplified).
--min-cluster-frequency FLOAT Low frequency cluster cutoff. [0.1]
--min-cluster-read-count INT Low read count cluster cutoff. [5]
--max-uchime-score FLOAT High UCHIME score cutoff. [1]
General Options:
--max-amplicon-size INT Upper read length cutoff, longer reads will be skipped. [15000]
--min-read-qv FLOAT Low read QV cutoff. [20]
-h,--help Show this help and exit.
--version Show application version and exit.
-j,--num-threads INT Number of threads to use, 0 means autodetection. [0]
--log-level STR Set log level. Valid choices: (TRACE, DEBUG, INFO, WARN, FATAL). [WARN]
--log-file FILE Log to a file, instead of stderr.
-h,--help Show this help and exit.
--version Show application version and exit.
-j,--num-threads INT Number of threads to use, 0 means autodetection. [0]
--log-level STR Set log level. Valid choices: (TRACE, DEBUG, INFO, WARN, FATAL). [WARN]
--log-file FILE Log to a file, instead of stderr.
```
### Coloring reads by clusters
If you have a BAM file (mapped amplicon reads) this tool will add IGV tags for grouping and coloring. It matches the read names in the clustering file, and the BAM file.
If you have a BAM file (mapped amplicon reads) this tool will add IGV tags for grouping (HP) and coloring (YC). It matches the read names in the clustering file, and the BAM file. The amplicon reads can be aligned against any reference.

```
pbaa bampaint - Add color tags to BAM records, based on pbaa clusters.
Expand All @@ -102,15 +109,15 @@ Options:

## Input

_pbaa_ requires two input files, guide/reference sequences in fasta file format and HiFi de-multiplexed reads in fastq file format. Guide/reference sequences should contain the amplified region, but not much more. Providing a chromosome, for example, will cause a large problems for placement.
_pbaa_ requires two input files, guide/reference sequences in fasta file format and HiFi de-multiplexed reads in fastq file format (indexed with fqidx). Guide/reference sequences should contain the amplified region, but not much more. Providing a chromosome as a guide, for example, will lead to issues during read placement.

_pbaa_ supports batching of samples via the FOFN (file of file name[s]) format. A FOFN is a line separated file that contains the **full paths** to the input files. **All input sequence files need to be indexed before running _pbaa_.** Indexing can be achieved with _samtools_ version 1.9 or greater.

## Customizing guide sequences

Guide/reference sequence choice affects read grouping/placement. It is important to choose guides that are sufficiently divergent. If too many similar alleles are used for the same locus the fraction of un-placed reads will increase because the number of informative kmers decrease within a locus. Too few guides can also cause cluster dropout; it's the goldilocks problem.
Guide/reference sequence choice affects read grouping/placement. It is important to choose guides that are sufficiently divergent. If too many similar alleles are used for the same locus the fraction of un-placed reads will increase because the number of informative guide kmers decreases. Too few guides can also cause cluster dropout; it's the goldilocks problem.

Guide sequences should be grouped into locus assignments. For example if multiple HLA-A alleles are used in the guide sequence, they should be grouped, so clustering will be performed at the locus level.
Guide sequences should be grouped into locus assignments. For example if multiple HLA-A alleles are used in the guide sequence, they should be grouped, so clustering will be performed at the locus level.

```
Allele_1|HLA-A (sequence name | group name)
Expand All @@ -137,13 +144,13 @@ The headers entries in the consensus sequence output contain statistics about th
example of a passing sequence:

```
>sample-bc1001--bc1001_guide-HLA-C_cluster-0_ReadCount-617 uchime_score:-1 uchime_left_parent: N/A uchime_right_parent: N/A cluster_freq:0.454345 diversity:0.0113006 avg_quality:87.0029 filters: none
>sample-bc1001--bc1001_guide-HLA-A_cluster-1_ReadCount-22 uchime_score:-1 uchime_left_parent:N/A uchime_right_parent:N/A cluster_freq:0.44 diversity:0.188552 avg_quality:53.5878 duplicate_parent:N/A seq_length:3152 filters:none
```

example of a failing sequence:

```
>sample-bc1001--bc1001_guide-HLA-A_cluster-6_ReadCount-32 uchime_score:3.33333 uchime_left_parent: bc1001--bc1001_HLA-A_0 uchime_right_parent: bc1001--bc1001_HLA-A_1 cluster_freq:0.0152164 diversity:2.30176 avg_quality:87.9363 filters: fail-high-chimera-score fail-cluster-is-low-frequency fail-high-diversity
>sample-bc1011--bc1011_guide-HLA-DRB1_cluster-3_ReadCount-6 uchime_score:0.03125 uchime_left_parent:bc1011--bc1011_HLA-DRB1_0 uchime_right_parent:bc1011--bc1011_HLA-DRB1_1 cluster_freq:0.12 diversity:0 avg_quality:42.2187 duplicate_parent:N/A seq_length:3706 filters:fail-low-frequency
```

The fields in the header are:
Expand All @@ -158,6 +165,8 @@ The fields in the header are:

5. **avg_quality** The average PHRED quality of the reads within the cluster.

6. **seq_length** The sequence length of the cluster.

6. **filters** This is a space separated field enumerating the possible reasons a cluster was placed in the fail category.

### Read Information Output File
Expand All @@ -168,21 +177,20 @@ One row per read, columns as follows:
4. SecondBestGuideName
5. Score
6. FirstHighest/SecondHighest/UniqueHitSum
7. Sample
8. VarString
9. ClusterId
10. ClusterProb
11. ClusterSize
12. ChimeraScore
7. Sample, input fastq
8. Sequence length
9. Average read quality
10. Cluster ID
11. Cluster Size

**_Example:_**
```
m54043_190914_194303/4195156/ccs HLA-B + HLA00158_B_14-02-01-01_4070_bp|HLA-B 0.663158 f:315/s:160/sum:619 reads.fasta bp-3300 0 1.000000 1 0.000000
m64012_200712_164638/72090819/ccs HLA-DRB5 - HLA00622_DQB1_02-01-01_7480_bp|HLA-DQB1 0.714286 f:5/s:2/sum:8 /pbi/dept/appslab/projects/old/2020/jh_hla/2020-07-13_HGgendx/fastq_sqII/demultiplex.bc1099--bc1099.fastq 3125 58.6149 1 1
```

## Best practices

### Sample preparation and sequencing
### Sample preparation and sequencing

[Targeted Sequencing For Amplicons Document](https://www.pacb.com/wp-content/uploads/Application-Brief-Targeted-sequencing-Best-Practices.pdf)

Expand All @@ -198,51 +206,39 @@ Amplification can generate off target reads. Similarly, pbaa can accidentally pl

A number of heuristics and advanced options are hidden from the interface. They are documented here, but consider them experimental features. Changing these settings can have undesired effects.

**_--no-indel_** : Only consider variants where both alleles are the same length in the bubble.

**_--skip-consensus_** : Only run read placement and clustering. No consensus sequences will be generated.

**_--no-hpc_** : Disable homopolymer compression for variant detection, and cluster filtering. Some variants can be masked by homopolymer compression.
**_--skip-chimera-detection_** : Skip chimera detection (UCHIME algorithm) step.

**_--kmer-size_** : Kmer size, not to exceed 31, length must be odd.
**_--kmer-size_** : Kmer size, not to exceed 31, length must be odd. This only affects read placement.

**_max-cluster-count_** : This flag interacts with _--fixed-cluster-count_, which must be set to zero to turn on this option. The algorithm attempts to learn the number of haplotype clusters. In practice a fixed upper limit runs quicker and generates very similar results.
## FAQ

**_min-path-coverage_** : Low coverage for bubble path cutoff. This options filters paths in the graph with fewer than N reads supporting the path at a fractional membership.
**_Why are reads missing_** : By default _pbaa_ only uses 500 reads per locus/guide/grouping. Increasing `--max-reads-per-guide` will use more reads, at the expense of runtime/memory. If runtime is not a consideration use all the reads.

## Demonstration Data
Demonstration data for testing __pbaa__ can be found [here](https://downloads.pacbcloud.com/public/dataset/pbAmpliconAnalysis_HLA/). Dataset contains HiFi reads for 6 pooled HLA genes in FASTQ format. Outputs from running _pbaa_ as well as validated genotypes are included.
**_Considerations for pooled experiments_**: Pooled experiments are bounded by runtime/memory. The settings below increase the accuracy of the results at the expense of runtime memory.

## FAQ
`--max-reads-per-guide` : Increase the number of reads uses locus/guide/grouping.

**_I'm finding extra false clusters_** : There are a number of reasons pbaa might generate false positive clusters. Chimeric reads, false variant calls, and clustering errors. Often the statistics in the fasta headers can provide clues. Changing filtering settings may reduce false positives.
`--max-alignments-per-read` : Increase the number of alignments per read. This setting can drastically increase computational expense.

**_I'm missing an expected cluster_** : _pbaa_ was tuned for sensitivity, favoring false positives over false negatives. However, false negatives do happen at a low rate. First check if the missing cluster is in the {prefix}_failed_cluster_sequences.fasta file. Then check that there is sufficient coverage over the guide sequences, by aligning reads to the guide sequences. For this application we recommend 50x depth of coverage per allele. If there is coverage check for missed variants. If you believe pbaa failed to discover a variant resulting in a false negative, please share a small example. Variant discover is the most challenging step in _pbaa_.
`--pile-size` : The number of reads to use for error masking per read.

**_I think I found a bug or I have feedback_** : Please share your experiences via github bug reports. Do not expect a quick reply.
`--min-var-frequency` : The minimum variant frequency within a pile of reads used for error correction. Decreasing this value will increase sensitivity to low-frequency variants. This can lead to over clustering.

## Disclaimer
THIS WEBSITE AND CONTENT AND ALL SITE-RELATED SERVICES, INCLUDING ANY DATA, ARE PROVIDED "AS IS," WITH ALL FAULTS, WITH NO REPRESENTATIONS OR WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTIES OF MERCHANTABILITY, SATISFACTORY QUALITY, NON-INFRINGEMENT OR FITNESS FOR A PARTICULAR PURPOSE. YOU ASSUME TOTAL RESPONSIBILITY AND RISK FOR YOUR USE OF THIS SITE, ALL SITE-RELATED SERVICES, AND ANY THIRD PARTY WEBSITES OR APPLICATIONS. NO ORAL OR WRITTEN INFORMATION OR ADVICE SHALL CREATE A WARRANTY OF ANY KIND. ANY REFERENCES TO SPECIFIC PRODUCTS OR SERVICES ON THE WEBSITES DO NOT CONSTITUTE OR IMPLY A RECOMMENDATION OR ENDORSEMENT BY PACIFIC BIOSCIENCES.
**_Extra false clusters_**: There are a number of reasons pbaa might generate false positive clusters. Chimeric reads, false variant calls, and clustering errors. Often the statistics in the fasta headers can provide clues. Changing filtering settings may reduce false positives.

**_Missing clusters_**: _pbaa_ was tuned for sensitivity, favoring false positives over false negatives. However, false negatives do happen at a low rate. First check if the missing cluster is in the {prefix}_failed_cluster_sequences.fasta file. Then check that there is sufficient coverage over the guide sequences, by aligning reads to the guide sequences. For this application we recommend a minimum of 25x depth of coverage per allele. If there is coverage check for missed variants. If you believe pbaa failed to discover a variant resulting in a false negative, please share a small example. Variant discover is the most challenging step in _pbaa_.

## Full Changelog
**_Feedback or bug report_**: Please share your experiences via [Github issue](https://github.com/PacificBiosciences/pbbioconda/issues).

* 0.1.4: March 25th, 2021
1. bugfix: Homopolymer compression can be turned off in duplicate marking using --no-hpc.
2. bugfix: Filter reads that are too short, before/after homopolymer compression.
## Disclaimer
THIS WEBSITE AND CONTENT AND ALL SITE-RELATED SERVICES, INCLUDING ANY DATA, ARE PROVIDED "AS IS," WITH ALL FAULTS, WITH NO REPRESENTATIONS OR WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTIES OF MERCHANTABILITY, SATISFACTORY QUALITY, NON-INFRINGEMENT OR FITNESS FOR A PARTICULAR PURPOSE. YOU ASSUME TOTAL RESPONSIBILITY AND RISK FOR YOUR USE OF THIS SITE, ALL SITE-RELATED SERVICES, AND ANY THIRD PARTY WEBSITES OR APPLICATIONS. NO ORAL OR WRITTEN INFORMATION OR ADVICE SHALL CREATE A WARRANTY OF ANY KIND. ANY REFERENCES TO SPECIFIC PRODUCTS OR SERVICES ON THE WEBSITES DO NOT CONSTITUTE OR IMPLY A RECOMMENDATION OR ENDORSEMENT BY PACIFIC BIOSCIENCES.

* 0.1.3: Feb 16, 2021
1. Updated CLI, highlighting faidx/fqidx required.

* 0.1.2:
1. Added BAM coloring binary.
## Full Changelog

* 0.1.1:
1. Simplification of the CLI.
2. Chimera detection across loci.
3. CLI now supports both single files and FOFNs.
4. Updated documentation.
* 1.0.0: April 23rd, 2021
1. Major algorithm refactor, and many changes to CLI.

* 0.1.0:
1. Beta code release.
2. Documentation.
* For previous releases please see [beta release documentation](README_BETA.md)
Loading

0 comments on commit 204078b

Please sign in to comment.