shlokanegi / GiraffeDeepVariant_Exome

Performance Evaluation of Giraffe-DeepVariant Workflow on WES data

GiraffeDeepVariant_Exome

Performance Evaluation of Giraffe-DeepVariant Workflow on WES data

Project Background

Variation Graph and Pangenome

Variation graphs provide a succinct encoding of the sequences of many genomes.
A Pangenome is a collection of genomes from individuals in a population and is represented as a variation graph.

Giraffe (Short-Read Mapping Algorithm)

Giraffe is a short-read mapping algorithm that finds the walk through the pangenomic graph that best matches the read.

Google’s DeepVariant (Variant Calling Algorithm)

Read alignments from Giraffe can be passed to DeepVariant, which visualizes these into pileup images (can be 6 or 7 channels deep) , and then uses a CNN to classify the images into variant calls.

Giraffe-DV v/s BWA-DV Performance Statistics by Google

Methodology

1. Conversion of BAM to paired-end fastQ files.
2. Attempted the GiraffeDeepVariant workflow on Terra.
3. Fixed the input parameter file to succeed.
4. Fetched Giraffe-PE VCF. Used it for hap.py benchmarking comparisons.
5. Investigations of exclusive BWA-only variants on IGV.

Results

hap.py Benchmarking Comparison

hap.py Benchmarking Results

IGV Visualizations

Conclusion and Future Work

• DeepVariant Model Retraining on correct paired-end exome data.
• Repeat analysis
• Evaluate performance of Giraffe on other kits, (Nextera and IDT) as well as the general ref-seq region.