DITTO

!!! For research purposes only !!!

NOTE: In a past life, DITTO used a different remote Git management provider, UAB Gitlab. It was migrated to Github in April 2023, and the Gitlab version has been archived.

DITTO is an explainable neural network that can be helpful for accurate and rapid interpretation of small genetic variants for pathogenicity using patient’s genotype (VCF) information.

Getting Started

• Prerequisites
• Using DITTO
  • Webapp
  • API
  • Prediction
    • Local Prediction
    • HPC Prediction with Cheaha
• Reproducing the DITTO model
• Download DITTO DB (Precomputed scores)
• How to cite?
• Contact

Prerequisites

The following prerequisites are required to be installed in the target envrionment for deploying and running DITTO prediction model.

Tools

• Python 3.10 - Install
  • The specified OpenCravat version requires Python 3.10
• Anaconda3 25.7+ - install
• OpenCravat 2.4.1 - install
• Git
  • Setup with your favorite git client. Here is a GitHub Guide for different platforms.
• Nextflow 22.10.7+ - install

NOTE: Current version of OpenCravat that we're using doesn't support "Spanning or overlapping deletions" variants i.e. variants with * in ALT Allele column. More on these variants

here.

These will be ignored when running the pipeline.

System Requirements

• CPU: >2
• RAM: ~25GB for a WGS VCF sample
• Storage: 1TB
  • The storage requirements are for hosting the OpenCravat annotators ~600GB of data required to store all annotators

Using DITTO

DITTO scores for variants can be obtained by the below 3 ways. Webapp and API are for single variant analysis and the local setup is for batch/bulk variant predictions.

Webapp

DITTO is available for public use at this website.

API

DITTO is not hosted as a public API but one can serve up locally to query DITTO scores. Please follow the instructions in this GitHub repo.

Prediction

Installation

To fetch DITTO source code, change in to directory of your choice and run:

git clone https://github.com/uab-cgds-worthey/DITTO.git
cd DITTO

Local Prediction

NOTE: This setup will allow one to annotate a VCF sample and make DITTO predictions. Currently tested only in Cheaha (UAB HPC) because of resource limitations to download datasets from OpenCRAVAT. Docker versions may need to be explored later to make it useable in Mac and Windows.

NextFlow Conda Vs. Mamba Setup

NOTE: If the user has conda running with Mamba instead of Conda, NextFlow can be configured to use Mamba instead by modifying the configs/nextflow/local.config file and updating the useMamba parameter to reflect the user's environment:

# This parameter is defaulted to false, change to true if using Mamba

useMamba = true

Setup Steps

• Setup OpenCravat (only one-time installation)

  Please follow the steps mentioned in install_openCravat.md.

• Setup Nextflow

  Create an environment via conda. Below is an example to install nextflow.

  # create environment. Needed only the first time. Please use the above link if you're not using Mac.
  conda env create -f ./configs/conda/ditto-env.yaml
  
  conda activate ditto-env

• Sample Sheet

  Please make a samplesheet .test_data/file_list.txt with VCF files (incl. path). One can supply either relative paths or absolute paths to files for the vcf.gz files. Relative paths need to be relative to the work directory that DITTO was executed from.

  Example file_list.txt with relative paths:

  .test_data/oc_test_data.vcf.gz
  .test_data/testing_variants_hg38.vcf.gz
  
  # Example, will become: /Users/<username>/Workspace/DITTO/.test_data/oc_test_data.vcf.gz

  Or absolute paths

  /Users/<username>/Desktop/test_data/oc_test_data.vcf.gz
  /Users/<username>/Desktop/test_data/testing_variants_hg38.vcf.gz
  
  # Example is using MacOS Desktop folder with test_data directory

  This will run DITTO prediction for both vcf files in the file_list.txt.

• Run the NextFlow pipeline

  Please make sure to edit the directory paths as needed and run the pipeline as shown below.

  # Note: NextFlow work directory is defined as `-work-dir` in the run command parameters
  # Note: `--output` cannot be relative, set a path nextflow can access. ex. `/tmp/DITTO/output`
  
  nextflow run pipeline.nf \
    -work-dir ./work_dir \
    --build hg38 -c ./configs/nextflow/local.config -with-report \
    --sample_sheet .test_data/file_list.txt \
    --oc_modules /<path-to>/opencravat/modules \
    --outdir $PWD/data/output

HPC Prediction with Cheaha

To run on UAB cheaha, see the installation step to clone the DITTO repository into a Cheaha directory.

• Create a text file listing the path to VCF file(s) (1 path per line) with variants to score

  • Paths can be full absolute paths or relative paths (relative to the directory where the pipeline will be run from, note the directory where the pipeline.nf file is)

• See the example input file .test_data/file_list.txt (lists 2 testing example input VCFs) for reference or as an input file for testing (default behavior of model.job)

  • One can supply either relative paths or absolute paths to files for the vcf.gz files. Relative paths need to be relative to the work directory that DITTO was executed from.

  Example file_list.txt with relative paths:

  .test_data/oc_test_data.vcf.gz
  .test_data/testing_variants_hg38.vcf.gz
  
  # Example, will become: /home/<username>/Workspace/DITTO/.test_data/oc_test_data.vcf.gz

  Or absolute paths

  /home/<username>/test_data/oc_test_data.vcf.gz
  /home/<username>/test_data/testing_variants_hg38.vcf.gz
  
  # Example is using Linux home directory with a test_data directory

• Update model.job (change the --sample_sheet option to your input file with VCF path(s) and --outdir to the desired output location of DITTO predictions)

sbatch model.job

Reproducing the DITTO model

Detailed instructions on reproducing the model is explained in build_DITTO.md

Download DITTO DB (Precomputed scores)

Precomputed scores for all possible SNVs and known Indels from gnomADv3.0 in main chromosomes in hg38 reference genome are available to download here - https://s3.lts.rc.uab.edu/cgds-public/dittodb/dittodb.html

How to cite?

Mamidi, T.K.K.; Wilk, B.M.; Gajapathy, M.; Worthey, E.A. DITTO: An Explainable Machine-Learning Model for Transcript-Specific Variant Pathogenicity Prediction. Preprints 2024, 2024040837. https://doi.org/10.20944/preprints202404.0837.v1

Contact information

For queries, please open a GitHub issue.

For urgent queries, send an email with clear description to

Name	Email
Tarun Mamidi	[email protected]
Liz Worthey	[email protected]