AhmedArslan / pepper

PEPPER-Margin-DeepVariant

P.E.P.P.E.R.

PEPPER is a genome inference module based on recurrent neural networks that enables long-read variant calling and nanopore assembly polishing in the PEPPER-Margin-DeepVariant pipeline. This pipeline enables nanopore-based variant calling with DeepVariant.

How to cite

Please cite the following manuscript if you are using PEPPER-Margin-DeepVariant:

Nature Methods: Haplotype-aware variant calling with PEPPER-Margin-DeepVariant enables high accuracy in nanopore long-reads.

Authors: Kishwar Shafin, Trevor Pesout, Pi-Chuan Chang, Maria Nattestad, Alexey Kolesnikov, Sidharth Goel,
Gunjan Baid, Mikhail Kolmogorov, Jordan M. Eizenga, Karen H. Miga, Paolo Carnevali, Miten Jain, Andrew Carroll & Benedict Paten.

Free access link to the manuscript: https://rdcu.be/cABfv

---

Long read variant calling performance evaluation

Detailed variant calling performance evaluation:

• ONT R9.4.1 Guppy 5.0.6 Sup:
  • HG003 whole genome (PEPPER vs Clair3).
  • HG002-HG006 chr20 (PEPPER vs Clair3).
• ONT R10.4 Q20 Guppy 5.0.15 Sup:
  • HG002 chr20 (PEPPER vs Clair3).
• PacBio HiFi:
  • HG003 chr20 (PEPPER-Margin-DeepVariant vs DeepVariant-WhatsHap-DeepVariant).

---

Useful links to documentations

• Quickstarts to check system configuration.
• Case-studies to reproduce performance.
• Description of PEPPER-Margin-DeepVariant method.
• How to train PEPPER-DeepVariant:
  • How to train PEPPER-SNP.
  • How to train PEPPER-HP.
  • How to train DeepVariant (using candidates from PEPPER).
• How to install PEPPER locally.
• List of parameters and description.
• How to tune parameters to balance speed and accuracy.

---

How to run

PEPPER-Margin-DeepVariant can be run using Docker or Singularity. A simple docker command looks like:

sudo docker run \
-v "${INPUT_DIR}":"${INPUT_DIR}" \
-v "${OUTPUT_DIR}":"${OUTPUT_DIR}" \
kishwars/pepper_deepvariant:r0.7 \
run_pepper_margin_deepvariant call_variant \
-b "${INPUT_DIR}/${BAM}" \
-f "${INPUT_DIR}/${REF}" \
-o "${OUTPUT_DIR}" \
-t "${THREADS}" \
--ont_r9_guppy5_sup

# --ont_r9_guppy5_sup is preset for ONT R9.4.1 Guppy 5 "Sup" basecaller
# for ONT R10.4 Q20 reads: --ont_r10_q20
# for PacBio-HiFi reads: --hifi

Quickstarts (small runs to test system configuration)

Test type	Links
Docker quickstart	Link
Singularity quickstart	Link
Docker-gpu quickstart	Link

Case studies (chromosome 20 runs for performance reproducibility)

Pipeline	Docker	Singularity	NVIDIA-docker (GPU)
Nanopore R9.4.1 variant calling	Link	Link	Link
Nanopore R10.4 Q20 variant calling	Link	Link	Link
PacBio HiFi variant calling	Link	Link	Link

License

PEPPER license, Margin License and DeepVariant License extend to the trained models (PEPPER, Margin and DeepVariant) and container environment (Docker and Singularity).

Acknowledgement

We are thankful to the developers of these packages:

• htslib & samtools
• pytorch
• ONNX
• hdf5 python (h5py)

Authors

PEPPER-Margin-DeepVariant pipeline is developed in a collaboration between UC Santa Cruz genomics institute and the Genomics team in Google Health.

Fun Fact

The name "P.E.P.P.E.R." is inspired from an A.I. created by Tony Stark in the Marvel Comics (Earth-616).

PEPPER is named after Tony Stark's then friend and the CEO of Resilient, Pepper Potts.