v1.2.0

1. Clair3 now natively supports GPU on Linux and Apple Silicon. Please refer to the GPU quick start guide for usage. Clair3 on GPU runs ~5 times compared CPU.

v1.1.2

1. Added boundary check where an insertion is immediately followed by a soft-clipping (#394, co-contributor @Devon Ryan)

2. Added exit code checking for all parallel jobs. The pipeline now immediately exits when encountering any job failure (#392, co-contributor @Sam Nicholls).

v1.1.1

1. Fixed the malformed VCF header issue that occurred specifically in AWS cloud environments(#380).
   2.Added a Clair3 R10.4.1 model fine-tuned on 12 bacterial genomes with improved variant calling performance for bacterial samples. Performance benchmarks and detailed results are documented in our note "fine-tuning_Clair3_with_12_bacteria_samples", (co-contributor @William Shropshire) .

v1.1.0

1. Removed parallel version checking (#377).
2. Bumped up minor version number in view of the new functions added in v1.0.11 #375.

v1.0.11

1. Added the --enable_variant_calling_at_sequence_head_and_tail option to enable variant calling at the head and tail 16bp of each sequence. Use with caution because alignments are less reliable in the regions, and there would be insufficient context to be fed to the neural network for reliable calling (#257
   by @0peterk87 and multiple contributors).
2. Added --output_all_contigs_in_gvcf_header to show all contigs IDs in the GVCF header (#371 by @alexnater), regardless whether a contig is covered at least a read or not.
3. Added a postprocessing script named AddPairEndAlleleDepth to add the "PEAD" tag for showing short-read pair-end allele depth (by Dr. Bin Guan from NEI).
4. Fixed AF value format issue in GVCF output that caused problem running bcftools merge (#365 by @jimmykhchiu). 5. Added a workflow to split reads into haplotypes according to phased variants and the do variant calling on single haplotypes. More details are in here.
5. Added set -o pipefail to run_clair3.sh so the non-zero returns of any part of the pipeline is now returned on failure (#368 by @Shians).
6. . Updated parameter descriptions for clarity (#369 by @Shians).

v1.0.10

1. Fixed an out of range bug when outputing GVCF for non-human genome (#317.
2. Improve the calling speed when working with targeted amplicon data, Use —chunk_num=-1 to call variant without splitting genome into chunks (#306, contributor Elliot Hallmark).
3. Updated LongPhase to version 1.7.3 (#321.

v1.0.9

1. Fixed an issue in VCF header(#305 by @Monica Palafox Roberts).
2. Updated DP FORMAT description in the header.

v1.0.8

1. Fixed an issue in VCF output that caused occasional quality score small differences compared to GVCF output.
2. Updated LongPhase to version 1.7.

v1.0.7

1. Added memory guards and test functions to the full-alignment calling C implementation (#286 by @Chris Wright).
2. Increased the maximum mpileup read coverage to 2^20 to accomodate excessively high-coverage amplicon data (#292 by @Devon Ryan).
3. Updated LongPhase to version 1.6.

v1.0.6

1. Fixed stack overflow issue when the read coverage is excessively high (#282 by @Chris Wright, #265 by @ymcki).
2. Added reference caching for CRAM input (#278 by @Alex Leonard, #180 by @bartcharbon and @SamStudio8).
3. Fixed a bug that outputs RefCall calls when no variant is called by full-alignment model (#271).
4. Fixed a bug that variants below min coverage were still being called (#262).
5. Set --min_snp_af and --min_indel_af to 0.0 when --vcf_fn is provided (#261).