VCF file generated from inference step is empty

[ypauling]

Hi deeppolisher team,
Thanks for developing this tool!
I tried to use this tool to polish degu genome built from PacBio hifi reads. After the inference step, the resulting vcf file is empty and the log file shows "Total window processed: 0". Below are some more details about how I used the tool.

The initial input files I have are two haplotype specific genomes (hap1 and hap2) and multiple hifi read files (fasta). Since we do not have ONT UL reads to run PHAROAPH pipeline, I combined the two haplotype genomes into a diploid genome and aligned the hifi reads to the diploid genome using minimap2 with default parameters. I combined bam files from multiple hifi read files and sorted and used the final merged bam file as the input for the deeppolisher tool.

I followed the case study to first generate the image directory, using the diploid genome and the merged sorted bam file as inputs. The output directory contains multiple *tfrecords.gz files which are ~1.5G each. I then used this directory and the checkpoint file provided in the docker image /opt/models/pacbio_model/checkpoint to run the inference step. The inference step finished in a few seconds, which might suggest that it cannot find any proper "window" in the provided diploid genome.

Do you know what might be the potential reasons for why the vcf file is empty? I can provide more details if that's helpful.

Thanks!
Bing

[kishwarshafin]

@ypauling it has been a while but can you paste the command here? Also the logs so it's easier to pinpoint what exactly went wrong.

[insilicool]

Empty VCF After DeepPolisher Inference

Hi,

I'm encountering an issue where the DeepPolisher inference step consistently produces an empty VCF file. This is happening for both my custom assembly and the provided case study data.

Problem Description
After successfully running polisher make_images, the subsequent polisher inference command completes without errors but generates an empty VCF.

Here are the commands I'm executing:

Case Study Data

micromamba activate deeppolisher_010 &&
polisher make_images --cpus 5
--bam input/HG002.trio_hifiasm_0.19.5.DC_1.2_40x.dip.minimap2v2.26.PHARAOH.chr20.bam
--fasta input/HG002.trio_hifiasm_0.19.5.DC_1.2_40x.pat.chr20.fasta
--output output/pat_images/

micromamba activate deeppolisher_010 &&
polisher inference --cpus 5
--input_dir output/pat_images/
--out_dir output/pat_vcf/
--checkpoint ${INSTALL_HOME}/software/micromamba/micromamba-2.0.2-0/envs/deeppolisher_010/share/models/pacbio_model/checkpoint
--reference_file input/HG002.trio_hifiasm_0.19.5.DC_1.2_40x.pat.chr20.fasta
--sample_name HG002

Custom Assembly (with Haplotagged BAM)

Initially, I noticed that the make_images --help suggested the BAM should be haplotagged, and the provided case study BAM did not appear to have HP tags. Consequently, I ran the same commands on my assembly of interest using a PHAROAH-generated, haplotagged diploid BAM. Unfortunately, I'm still getting the same result: an empty VCF.

micromamba activate deeppolisher_010 &&
polisher make_images --cpus 15
--bam verkko221_HiFiSPRQ_ONTUL_HIC.hifi.haplotagged.bam
--fasta verkko221_HiFiSPRQ_ONTUL_HIC.haplotype1.fasta
--output deeppolisher/haplotype1/haplotype1

micromamba activate deeppolisher_010 &&
polisher inference --cpus 15
--input_dir deeppolisher/haplotype1/haplotype1
--out_dir vcf/haplotype1/
--checkpoint ${INSTALL_HOME}/software/micromamba/micromamba-2.0.2-0/envs/deeppolisher_010/share/models/pacbio_model/checkpoint
--reference_file verkko221_HiFiSPRQ_ONTUL_HIC.haplotype1.fasta
--sample_name HG002

Request for Assistance

Could you please provide some insight into what might be causing this puzzling issue? Any guidance on troubleshooting steps or potential misconfigurations would be greatly appreciated.

Thanks.

Best,
Rob E.