proovframe: frameshift-correction for long-read (meta)genomics

Abstract

Long-read sequencing technologies hold big promises for the genomic analysis of complex samples such as microbial communities. Yet, despite improving accuracy, basic gene prediction on long-read data is still often impaired by frameshifts resulting from small indels. Consensus polishing using either complementary short reads or to a lesser extent the long reads themselves can mitigate this effect but requires universally high sequencing depth, which is difficult to achieve in complex samples where the majority of community members are rare. Here we present proovframe, a software implementing an alternative approach to overcome frameshift errors in long-read assemblies and raw long reads. We utilize protein-to-nucleotide alignments against reference databases to pinpoint indels in contigs or reads and correct them by deleting or inserting 1-2 bases, thereby conservatively restoring reading-frame fidelity in aligned regions. Using simulated and real-world benchmark data we show that proovframe performs comparably to short-read-based polishing on assembled data, works well with remote protein homologs, and can even be applied to raw reads directly. Together, our results demonstrate that protein-guided frameshift correction significantly improves the analyzability of long-read data both in combination with and as an alternative to common polishing strategies. Proovframe is available from https://github.com/thackl/proovframe.

Oxford Nanopore R10.4 long-read sequencing enables the generation of near-finished bacterial genomes from pure cultures and metagenomes without short-read or reference polishing

“…A commonly adopted solution has been to include short-read data for post-assembly error correction 15,22 , although it increases the cost and complexity overhead. Another solution has been to apply reference-based polishing to correct frameshift errors [23][24][25] but, although this provides a practical solution that enables gene calling, it does not provide true near-finished genomes. Finished microbial genomes, as defined by Bowers et al 2017 in the MIMAG (minimum information about a metagenome-assembled genome) standard 26 , are genomes that have "...a single, validated, contiguous sequence per replicon, without gaps or ambiguities" and "a consensus error rate equivalent to Q50 or better".…”

Oxford Nanopore R10.4 long-read sequencing enables near-perfect bacterial genomes from pure cultures and metagenomes without short-read or reference polishing

“…A commonly adopted solution has been to include short-read data for post-assembly error correction 12,19 , although it increases the cost and complexity overhead. Another solution has been to apply reference-based polishing to correct frameshift errors [20][21][22] , but while it provides a practical solution, which allows gene calling, it does not provide true near-perfect genomes.…”

BugSplit: highly accurate taxonomic binning of metagenomic assemblies enables genome-resolved metagenomics

“…Full details are available in the original PointFinder 61 methods. We base our modifications to PointFinder on the previously demonstrated observation that frameshifts and stop codons in third-generation assemblies are more likely to reflect sequencing and assembly errors than true sequence variation 62,63 . We modify PointFinder to not halt its search for variants along a resistance loci if it encounters a stop codon.…”

“…We modify PointFinder to not halt its search for variants along a resistance loci if it encounters a stop codon. We additionally modify PointFinder to shift alignments around indels, maintaining the reading frame, in an approach similar to more general frameshift correction tools 62,63 . Our modified PointFinder has been incorporated into ResFinder version 4.2 and can be activated with the ‘-ii’ (Ignore Indels) and ‘-ic’ (Ignore stop Codons) flags.…”