Oxford_nanopore_r9_whole_genome.md

ONT R9.4.1 Guppy 5.0.7 "Sup" HG003 whole genome performance evaluation

We hold out HG003 sample while training PEPPER-Margin-DeepVariant so we use HG003 to demonstrate our whole genome performance. We report both runtime and accuracy for this evaluation between PEPPER-Margin-DeepVariant r0.7 and r0.8.

Setup

We used the following dataset:

Sample:                   HG003 (Whole genome)
Coverage:                 ~85x
Chemistry:                R9.4.1
Basecaller:               Guppy 5.0.7 "Sup"

Downsampling the alignment file:

We downsampled the ~85x variant calling data using the following command:

samtools view -s 0.71 -b -@${THREADS} HG003_guppy_507_2_GRCh38_pass.bam > HG003_guppy_507_2_GRCh38_pass.60x.bam
samtools view -s 0.36 -b -@${THREADS} HG003_guppy_507_2_GRCh38_pass.bam > HG003_guppy_507_2_GRCh38_pass.30x.bam

Calling variants with PEPPER-Margin-DeepVariant (PEPPER r0.8)

time docker run -it -v /data:/data \
-u `id -u`:`id -g` \
kishwars/pepper_deepvariant:r0.8 \
run_pepper_margin_deepvariant call_variant \
-b $BAM \
-f $REF \
-o $OUTPUT_DIR \
-t $THREADS \
-s HG003 \
--ont_r9_guppy5_sup 2>&1 | tee $LOG_FILE

Results

In all stratified coverages (30x, 60x, 85x), PEPPER-Margin-DeepVariant r0.8 shows increased accuracy:

HG003 30x performance:

Sample	Version	Type	Truth total	True positives	False negatives	False positives	Recall	Precision	F1-Score
HG003 30x	r0.7	INDEL	504501	317621	186880	35084	0.629575	0.902714	0.7418
		SNP	3327495	3310002	17493	11986	0.994743	0.996393	0.995567
	r0.8	INDEL	504501	345384	159117	51842	0.684605	0.872481	0.767209
		SNP	3327495	3309038	18457	9173	0.994453	0.997236	0.995843

HG003 60x performance:

Sample	Version	Type	Truth total	True positives	False negatives	False positives	Recall	Precision	F1-Score
HG003 60x	r0.7	INDEL	504501	366144	138357	33484	0.725755	0.91827	0.810741
		SNP	3327495	3317492	10003	8548	0.996994	0.99743	0.997212
	r0.8	INDEL	504501	394987	109514	44678	0.782926	0.90091	0.837785
		SNP	3327495	3317515	9980	7120	0.997001	0.997859	0.99743

HG003 85x performance:

Sample	Version	Type	Truth total	True positives	False negatives	False positives	Recall	Precision	F1-Score
HG003 85x	r0.7	INDEL	504501	383384	121117	30595	0.759927	0.927982	0.835588
		SNP	3327495	3318437	9058	8032	0.997278	0.997586	0.997432
	r0.8	INDEL	504501	412169	92332	38633	0.816984	0.91651	0.86389
		SNP	3327495	3318308	9187	6733	0.997239	0.997976	0.997607