Portello: Making global assembly more effective for rare-disease WGS

Presented at CSHL Genome Informatics 2025.

Original Abstract:

We introduce portello, a new method to translate HiFi read alignments from diploid assembly contigs to a standard reference genome, such as GRCh38. This translation process only requires an existing alignment of the assembly contigs to the reference. We demonstrate that these translated read alignments are superior to those from conventional mapping, particularly in regions of large-scale variation. Portello translated alignments also provide a more effective method for read-backed phasing by leveraging each read’s association to an assembly contig, which can be used to phase haplotypes even through complex structural variation. We additionally demonstrate that portello read alignments help to improve small variant calling accuracy, due in part to improving the consistency of indel alignments. For DeepVariant calls made from portello alignments, small variant false negatives and false positives are reduced by 41.4% and 5.6%, respectively, compared to DeepVariant calls from conventional read mapping with pbmm2 (assessment on NA12878, using Platinum Pedigree small variant benchmark v1.2). Notably, these results reflect the standard DeepVariant v1.9 HiFi model without any retraining. This outcome demonstrates that portello provides a way to unify methods from assembly and conventional mapping approaches into a single assembly-based workflow. Such a workflow, providing both the sample’s personal diploid assembly as well as read-to-reference alignments translated from read-to-assembly alignments, allows methods written for conventional mapping approaches to coherently operate together with assembly-based variant calling and analysis. This approach should be especially valuable for human rare-disease analysis where long-read assembly is routinely used to improve detection and characterization of denovo variants in the proband.