These documents refer to an obsolete way of installing and running FALCON. They will remain up for historical context and for individuals still using the older version of FALCON/FALCON_unzip.


The current PacBio Assembly suite documentation which includes new bioconda instructions for installing FALCON, FALCON_unzip and their associated dependencies can be found here pb_assembly

FALCON Assembler


In this section we will run the FALCON pipeline on an E. coli dataset. We will work through the commands and results and give you ideas of how to assess the perfomance of FALCON on your dataset so you can modify parameters and trouble-shoot more effectively. This tutorial is a beginners guide to FALCON but assumes bioinformatics fluency.

Input Files

You will need three types of files to get started, your PacBio data in fasta format (can be one or many files), a text file telling FALCON where to find your fasta files, and a configuration file. All files except the fasta files must be in your job directory.

1. Download E. coli dataset

An example E. coli dataset can be download from here. and then unpacked. e.g.:

$ wget
$ tar -xvzf ecoli.m140913_050931_42139_c100713652400000001823152404301535_s1_p0.subreads.tar.gz

You should find three fasta files of ~350 Mb each in the newly created subdirectory.

2. Create FOFN

Next, create a “file-of-file-names”, (“fofn”) with the full path of each fasta file, one per line.


3. Download configuration file

If you are running on a cluster with a scheduler use this a starting point: fc_run_ecoli.cfg If you are running your job locally, try this file: fc_run_ecoli_local.cfg

These config files are meant to be starting points only! You will need to make adjustments according to your particular compute setup. Please refer to these config files as you plan your run.

Note that I have manually specified a seed read length cutoff of 15kb rather than using an automated cut off (length_cutoff = -1, with calculated length cut off = 22486). The raw read coverage is very high (>200X); by reducing the seed read length cutoff, we avoid enriching our seed reads for erroneous chimeric (and very long) reads. Try running the assembly using the automated seed read length cut off, you should get a fragmented (28 contigs) and incomplete assembly (< 900Mb).

Running FALCON

I send all of my FALCON jobs to the scheduler for ease of tracking job progress. Here is an example bash script that submits to an SGE cluster:

#$ -S /bin/bash
#$ -N myJob
#$ -cwd
#$ -q myqueue

# load dependencies
module load python/2.7.9 gcc/4.9.2

# source build
cd /path_to_build/src/FALCON-integrate/

# navigate to job directory, directory containing input.fofn
cd /path/to/my/job_dir

# run it!
fc_run fc_run.cfg

To initiate the FALCON run, I just submit my job to the scheduler with a qsub command:

$ qsub

Alternatively, you can add the fc_env/bin directory to your $PATH and invoke fc_run at the command line with your fc_run.cfg as the argument. Note that this shell needs to persist through the entire assembly process so you may need to use a window manager like screen to maintain your connection.

falcon_jobdir$ export PYTHONUSERBASE=/path_to_build/fc_env/
falcon_jobdir$ export PATH=$PYTHONUSERBASE/bin:$PATH
falcon_jobdir$ fc_run fc_run.cfg

Assessing Run Progress

Refer to the pipeline document for detailed summary of FALCON job directory structure, sequence of commands, and output files created.

Counting Completed Jobs

The majority of run-time is spent during the daligner phases, performing the alignments and then sorting and merging them. To determine how many jobs are performed for each step, refer to 0-rawreads/

$ grep '^#' 0-rawreads/

    # Daligner jobs (60)
    # Initial sort jobs (400)
    # Check initial .las files jobs (80) (optional but recommended)
    # Remove initial .las files
    # Level 1 merge jobs (20)
    # Check level 2 .las files jobs (20) (optional but recommended)
    # Remove level 1 .las files (optional)

To determine how many jobs have completed, count the sentinel files that indicate a job is complete. For example:

$ find 0-rawreads/ -name "job*done" | wc -l

$ find 0-rawreads/ -name "m_*done" | wc -l

Assessing Run Performance

Raw and Pread Coverage and Quality

The E. coli subreads are a total of 1.01 Gb of data in 105,451 reads. is a useful script by Joseph Fass and Brad Sickler at UC Davis for calculating total sequence length and other assembly metrics).

You can confirm that your dazzler database was correctly constructed using a utility from the dazzler suite:

$ DBstats raw_reads.db > raw_reads.stats
$ head raw_reads.stats -n 17

Statistics for all reads of length 500 bases or more

        90,747 reads        out of         105,451  ( 86.1%)
        964,281,429 base pairs   out of   1,013,118,375  ( 95.2%)

        10,626 average read length
        6,805 standard deviation

Base composition: 0.248(A) 0.242(C) 0.263(G) 0.246(T)

Distribution of Read Lengths (Bin size = 1,000)

        Bin:      Count  % Reads  % Bases     Average
        45,000:     1      0.0      0.0       45611

You can see that we discarded 13.9% of the raw bases and 4.8% of the reads by employing a raw read length cut off of 500bp in the DBsplit options. This file can also be used to plot a histogram of raw read lengths.

The genome of this E. coli strain is 4.65 Mb long for a raw read coverage of ~207 fold. Confirm this with the preassembly report:

$ cat 0-rawreads/report/pre_assembly_stats.json

"genome_length": 4652500,
"length_cutoff": 15000,
"preassembled_bases": 350302363,
"preassembled_coverage": 75.293,
"preassembled_mean": 10730.33,
"preassembled_n50": 16120,
"preassembled_p95": 22741,
"preassembled_reads": 32646,
"preassembled_seed_fragmentation": 1.451,       # number split preads / seed reads
"preassembled_seed_truncation": 4453.782,       # ave bp lost per pread due to low cov
"preassembled_yield": 0.758,                    # total pread bp / seed read bp
"raw_bases": 964281429,
"raw_coverage": 207.261,
"raw_mean": 10626.042,
"raw_n50": 14591,
"raw_p95": 23194,
"raw_reads": 90747,
"seed_bases": 461851093,
"seed_coverage": 99.269,                        # raw base coverage depth on seed reads
"seed_mean": 20029.103,
"seed_n50": 19855,
"seed_p95": 28307,
"seed_reads": 23059

A note on these statistics: in the process of created preads, seeds reads with insufficient raw read coverage (usually due to base errors) will be split or truncated. The preassembled seed fragmentation, truncation, and yield stats summarize the quality of pread assembly. A good preassembled yield should be greater than 50%. Note that if an automated seed read length is used for this data (22486), preassembled seed read truncation is ~6kb, indicating that many of the longest raw reads are not supported by the rest of the data.

You can similarly summarize the contents of the dazzler database for preads using DBstats and plotting in R.

Contig Stats

When your run is complete, you can summarize your assembly stats using the script:

$ p_ctg.fa > p_ctg.stats
$ a_ctg.fa > a_ctg.stats
$ tail p_ctg.stats

Total length of sequence:       4635395 bp
Total number of sequences:      1
N25 stats:                      25% of total sequence length is contained in the 1 sequences >= 4635395 bp
N50 stats:                      50% of total sequence length is contained in the 1 sequences >= 4635395 bp
N75 stats:                      75% of total sequence length is contained in the 1 sequences >= 4635395 bp
Total GC count:                 2352187 bp
GC %:                           50.74 %

Assembly Graph and Pread Overlaps

Assembly contiguity can be enhanced by adjusting a few parameters in the last stage of the assembly process. You can try a grid of pread length cut offs for the filtering of the final overlaps in the assembly graph. In a general sense, longer pread length cut offs will increase the contiguity (contig N50) in your assembly, but may result in shorter over all assembly length. To try different length cut off, rename your 2-asm-falcon dir, modify the config file, rename the log and mypwatcher directory, and restart FALCON:

$ mv 2-asm-falcon 2-asm-falcon_12kb
$ mv mypwatcher/ mypwatcher0/
$ mv all.log all0.log
$ qsub

The other parameter to adjust is the number of overlaps in the assembly graph. First, look at a histogram of the number of overlaps on the 5’ and 3’ end of each read. Run the falcon utility:

# make sure utility is in $PATH
$ export PYTHONUSERBASE=/path_to_build/fc_env/

# navigate to directory
$ cd 2-asm-falcon
$ fc_ovlp_stats --fofn ../1-preads_ovl/merge-gather/las.fofn > ovlp.stats

Then plot histograms of the number of 5’ and 3’ overlaps between preads in R. This can inform your parameters for sge_option_fc where min_cov and max_cov should flank the bulk of the distribution. For repetative genomes, a second mode in the distribution may appear, containing preads ending or begining in repetative material. It is best to choose a max_cov to the left of the repeat mode that removes these repetative overlaps.

Troubleshooting Run

If you find your run has died here are some suggestions of how to restart, in order of increasing difficulty:

Simple Restart

Simply rename your log file and mypwatcher directory and restart the pipeline. Renaming these files preserves them for you reference, and by removing the original mypwatcher directory the pipeline, when restarted, will scan your job directory for completed jobs and pick up where it left off:

$ mv mypwatcher/ mypwatcher0/
$ mv all.log all0.log
$ qsub

Directory Cleanup and Restart

First, determine which job caused the run to fail. For example:

$ grep 'ERROR' all.log

2016-11-21 03:21:39,482 - pypeflow.simple_pwatcher_bridge - ERROR - Task Node(0-rawreads/m_00210) failed with exit-code=99
2016-11-21 03:21:39,482 - pypeflow.simple_pwatcher_bridge - ERROR - Failed to clean-up FakeThread: jobid=Pcfbdb8b3c50d5e status='EXIT '

Delete all directories that failed, then rename the log file and mypwatcher as above:

$ rm -rf 0-rawreads/m_00210
$ mv mypwatcher/ mypwatcher0/
$ mv all.log all0.log
$ qsub

You can find out more details about the failed jobs in mypwatcher/ to diagnose the problem.

$ less mypwatcher/jobs/Pcfbdb8b3c50d5e/stderr
$ less mypwatcher/jobs/Pcfbdb8b3c50d5e/stdout

Manual Running of Failed Jobs

If your job still fails, try manually running the problematic jobs. Search in the job directory for the shell script containing the individual tasks and try manually running the shell script or individual tasks:

$ ls job_0000

job_0000_done  L1.19.5.las  L1.19.7.las  L1.5.19.las  L1.7.19.las   raw_reads.db       task.json
L1.19.4.las    L1.19.6.las  L1.4.19.las  L1.6.19.las  pwatcher.dir

$ head job_0000/ -n 12

set -vex

ln -sf ${db_dir}/.raw_reads.bps .
ln -sf ${db_dir}/.raw_reads.idx .
ln -sf ${db_dir}/raw_reads.db .
ln -sf ${db_dir}/.raw_reads.dust.anno .
ln -sf ${db_dir}/ .
daligner -v -t16 -H22486 -e0.7 -s1000 raw_reads.19 raw_reads.4 raw_reads.5 raw_reads.6 raw_reads.7
LAcheck -v raw_reads *.las
LAsort -v raw_reads.4.raw_reads.19.C0 raw_reads.4.raw_reads.19.N0 raw_reads.4.raw_reads.19.C1 raw_reads.4.raw_reads.19.N1 raw_reads.4.raw_reads.19.C2 raw_reads.4.raw_reads.19.N2 raw_reads.4.raw_reads.19.C3 raw_reads.4.raw_reads.19.N3 && LAmerge -v L1.4.19 raw_reads.4.raw_reads.19.C0.S raw_reads.4.raw_reads.19.N0.S raw_reads.4.raw_reads.19.C1.S raw_reads.4.raw_reads.19.N1.S raw_reads.4.raw_reads.19.C2.S raw_reads.4.raw_reads.19.N2.S raw_reads.4.raw_reads.19.C3.S raw_reads.4.raw_reads.19.N3.S

Once these jobs have run to completion, you can try restarting the pipeline.