Improved machine learning / improved variant comparison / new SV comparison tools
March 22, 2018
Real Time Genomics are pleased to announce the availability of new releases of our full analysis suite, RTG Core, and our utility package, RTG Tools. This release includes new features and performance improvements. Some of the highlights of this release:
- Representation improvements to variant caller outputs. The various variant calling commands default to an alternative algorithm for representing haplotype calls as smaller components. While the underlying haplotype calls are the same as previous releases, the decomposed representation used in VCF is now more granular, so results can look quite different on the surface.
- Small-variant evaluation improvements. vcfeval allows optional preprocessing of input VCF files to decompose large calls to smaller constituents, which can permit longer calls to receive partial credit during accounting. In addition, vcfeval snp and indel ROC outputs now include precision and sensitivity metrics. For more information see the user manual. vcfeval also supports matching variants that occur inside a spanning deletion.
- Support for structural variant evaluation. This release includes beta commands for comparing structural variant calls such as translocations, inversions, and sequence-resolved larger insertions and deletions, using a similar workflow to vcfeval/rocplot. The new svdecompose command converts higher-level SV events and longer insertions and deletions into break-ends. In conjunction with this, the new bndeval command runs a comparison between a baseline and called break-end dataset. Outputs are VCF and ROC data files that are compatible with rocplot.
- Java 9 compatibility testing. RTG is compatible with Java 9, although currently we recommend Java 8 for performance reasons. Also note that due to differences in Java Math library implementation between Java 8 and Java 9, in rare situations minor output differences may be observed when comparing results obtained using Java 8 with Java 9. Builds that include a bundled JRE have been updated to the latest JRE 8u161.
- Improvements to the AVR models that perform variant scoring. The variant callers include new predictor attributes and all AVR models have been rebuilt to take advantage of these.