Learning through An Example#

We will walk through a quick end-to-end flow for testing and debugging a variant calling workflow to demonstrate how to use latch develop.

Prerequisites#

Building a Simple Variant Calling Workflow#

In this tutorial, we will be building a variant calling workflow. To follow along, clone the example code here:

git clone https://github.com/hannahle/simple-variant-calling.git
cd simple-variant-calling

The repository consists of three folders:

  • buggy-wf: The workflow with bugs

  • good-wf: The final, functional workflow

  • wgs: Test data for the workflow

We will use latch develop to test and debug the buggy-wf to arrive at good-wf effectively.

Let’s get started!

Preparing test data#

First, we must upload our test data folder, wgs, to the Latch Platform.

Run the following command to upload the wgs data from the terminal

$ latch cp wgs latch:///wgs

Once the data has finished uploading, verify whether it exists on Latch by using latch ls like so:

$ latch ls

Size Date Modified Name
   - -             welcome/
   - -             wgs/

Overview of the variant calling workflow#

The data we are working with is part of a long-term evolution experiment by Richard Lenski, designed to assess the adaptation of E. coli in various environments.

Variant calling is a typical workflow used to observe the change in a population over successive generations. We can use this to analyze how this experiment’s E. coli_ population changed over time relative to the original population, E. coli strain REL606. To do so, we will align each of our samples to the original E.coli strain’s (REL606) reference genome to determine the differences between our reads after 40,000 generations versus the original genome.

Upload

Our variant calling pipeline will consist of five steps:

  1. Index the reference genome

  2. Align our reads to the reference genome

  3. Convert our aligned reads from SAM to BAM format

  4. Sort our BAM file by coordinates

  5. Perform variant calling

Hence, our workflow will contain five tasks: build_index, align_reads, convert_to_bam, sort_bam, and variant_calling.

We have provided some (buggy) code for the five tasks above for you in the buggy-wf, which you will now test and debug!

Testing and Debugging the Workflow#

First, enter the buggy-wf folder and register it to Latch:

$ ls
buggy-wf        good-wf         wgs

$ cd buggy-wf

$ latch register --remote .

We must register the workflow before we can debug it with latch develop. The registration process builds the code’s environment, which is the key to successfully debugging your workflow. Now we can interact with the environment.

Entering the environment#

Run latch develop . in the workflow directory to enter the workflow environment.

We recommend reading this overview of the latch develop environment before proceeding.

Defining a test script#

Before testing the workflow end-to-end, it is helpful to run and test each task individually. To get started, we have provided a testing directory called scripts with a main.py inside containing commented-out task test code.

$ tree .
.
├── Dockerfile
├── README.md
├── scripts
│   └── main.py
├── version
└── wf
    └── __init__.py

3 directories, 6 files

Uncomment some of the code in main.py to get this:

# Import task functions from our workflow
from wf import build_index, align_reads, convert_to_bam, sort_bam, variant_calling
from latch.types import LatchFile, LatchDir

# Call task function
build_index(ref_genome = LatchFile("latch:///wgs/ref_genome/ecoli_rel606.fasta"))
  • The first line imports all tasks defined in wf/__init__.py so we can reference them in this script.

  • The second line imports the necessary Latch types.

  • The third line calls the task function to index the reference genome, build_index.

Copy

To use a file on Latch as test data for the task, navigate to the Latch Console, click on the specific file and copy the path shown on the sidebar. After copying the path, prefix it with latch:// to specify that it is a file on Latch, and pass the whole string as a parameter to LatchFile.

Calling the test script#

Now that we have modified our local code, we can run it in the development environment. Note that local changes are automatically reflected in the development environment.

>>> python3 scripts/main.py

...
FileNotFoundError: [Errno 2] No such file or directory: 'bwa'

The logs tell us that there is no file or directory called bwa. One potential reason is that we might not have installed the binary bwa correctly.

>>> bwa

bash: bwa: command not found

Indeed, our bwa binary is not installed! Checking the Dockerfile, notice that the installation instruction for bwa is commented out. Let’s uncomment it:

...
RUN apt-get install bwa
...

Because we modified our Dockerfile, we must rebuild the environment and enter a new development session to load in the newest changes. First, exit the current development session:

>>> exit
Exiting local development session

Re-register the workflow with the new Docker image

$ latch register --remote .
...

Now enter a new development session and re-run the test script:

>>> python3 scripts/main.py

The script should now run successfully!

Where are my outputs?#

To ensure that our tasks are working correctly, let’s look at their output files to ensure they’re correct. To locate the output files, we can inspect the return statement of the build_index task inside wf/__init__.py:

@small_task
def build_index(ref_genome: LatchFile = LatchFile("latch:///wgs/ref_genome/ecoli_rel606.fasta")) -> LatchDir:
    _bwa_cmd = [
        "bwa",
        "index",
        ref_genome.local_path,
    ]
    subprocess.run(_bwa_cmd)
    output = Path(ref_genome.local_path).resolve().parent

    return LatchDir(output, "latch:///wgs/ref_genome")

We see that the task is returning a LatchDir with the remote path latch:///wgs/ref_genome, which indicates that the output files are inside the /wgs/ref_genome folder in the Latch Console.

Outputs

Debugging Subsequent Tasks#

Similarly to how we tested our first task, we can also call the second task from scripts/main.py like so:

from wf import build_index, align_reads, convert_to_bam, sort_bam, variant_calling
from latch.types import LatchFile, LatchDir

build_index(ref_genome=LatchFile("latch:///wgs/ref_genome/ecoli_rel606.fasta"))

align_reads(
    ref_genome_dir=LatchDir("latch:///wgs/ref_genome/"),
    read1=LatchFile("latch:///wgs/data/SRR2584863_1.trim.sub.fastq"),
    read2=LatchFile("latch:///wgs/data/SRR2584863_2.trim.sub.fastq"),
)

Here, we are passing the output of the first task, latch:///wgs/ref_genome, as the input to the second task.

Run the test script:

>>> python3 scripts/main.py
...
Original exception: failed to get presigned url for `latch:///wgs/trimmed_fastqs/SRR2584863_1.trim.sub.fastq`

This error tells us that there is no file called /wgs/trimmed_fastqs/SRR2584863_1.trim.sub.fastq in the Latch Console. Referencing the actual file paths of the trimmed FastQs, we can see that their paths are indeed wrong, with the correct paths being /wgs/data/SRR2584863_1.trim.sub.fastq, etc (listed under /wgs/data instead of /wgs/ trimmed_fastqs).

We can make this modification to our test script and re-run the task as below:

>>> python3 scripts/main.py

The task now outputs the results to the folder /results on Latch!

Aligned

Exercise#

As an optional exercise, continue debugging the workflow’s final three tasks. The debugged workflow code is provided in the good-wf folder for reference.


Key Takeaways#

  • How to open a development session with latch develop

  • How local changes are synced to the development environment

  • How to reflect changes from your Dockerfile in the latch develop environment