BACKGROUND: The accurate detection of variants is essential for genomics-based studies. Currently, there are various tools designed to detect genomic variants, however, it has always been a challenge to decide which tool to use, especially when various major genome projects have chosen to use different tools. Thus far, most of the existing tools were mainly developed to work on short-read data (i.e., Illumina); however, other sequencing technologies (e.g. PacBio, and Oxford Nanopore) have recently shown that they can also be used for variant calling. In addition, with the emergence of artificial intelligence (AI)-based variant calling tools, there is a pressing need to compare these tools in terms of efficiency, accuracy, computational power, and ease of use. RESULTS: In this study, we evaluated five of the most widely used conventional and AI-based variant calling tools (BCFTools, GATK4, Platypus, DNAscope, and DeepVariant) in terms of accuracy and computational cost using both short-read and long-read data derived from three different sequencing technologies (Illumina, PacBio HiFi, and ONT) for the same set of samples from the Genome In A Bottle project. The analysis showed that AI-based variant calling tools supersede conventional ones for calling SNVs and INDELs using both long and short reads in most aspects. In addition, we demonstrate the advantages and drawbacks of each tool while ranking them in each aspect of these comparisons. CONCLUSION: This study provides best practices for variant calling using AI-based and conventional variant callers with different types of sequencing data.

MOTIVATION: Identification of DNA sequence variations such as single nucleotide polymorphisms (SNPs) is a fundamental step toward genetic studies. Reduced-representation sequencing methods have been developed as alternatives to whole genome sequencing to reduce costs and enable the analysis of many more individual. Amongst these methods, restriction site associated sequencing (RSAS) methodologies have been widely used for rapid and cost-effective discovery of SNPs and for high-throughput genotyping in a wide range of species. Despite the extensive improvements of the RSAS methods in the last decade, the estimation of the number of reads (i.e. read depth) required per sample for an efficient and effective genotyping remains mostly based on trial and error. RESULTS: Herein we describe a bioinformatics tool, DepthFinder, designed to estimate the required read counts for RSAS methods. To illustrate its performance, we estimated required read counts in six different species (human, cattle, spruce budworm, salmon, barley and soybean) that cover a range of different biological (genome size, level of genome complexity, level of DNA methylation and ploidy) and technical (library preparation protocol and sequencing platform) factors. To assess the prediction accuracy of DepthFinder, we compared DepthFinder-derived results with independent datasets obtained from an RSAS experiment. This analysis yielded estimated accuracies of nearly 94%. Moreover, we present DepthFinder as a powerful tool to predict the most effective size selection interval in RSAS work. We conclude that DepthFinder constitutes an efficient, reliable and useful tool for a broad array of users in different research communities. AVAILABILITY AND IMPLEMENTATION: https://bitbucket.org/jerlar73/DepthFinder. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Reduced-representation sequencing is a genome-wide scanning method for simultaneous discovery and genotyping of thousands to millions of single nucleotide polymorphisms that is used across a wide range of species. However, in this method a reproducible but very small fraction of the genome is captured for sequencing, while the resulting reads are typically aligned against the entire reference genome. RESULTS: Here we present a skinny reference genome approach in which a simplified reference genome is used to decrease computing time for data processing and to increase single nucleotide polymorphism counts and accuracy. A skinny reference genome can be integrated into any reduced-representation sequencing analytical pipeline. AVAILABILITY AND IMPLEMENTATION: https://bitbucket.org/jerlar73/SRG-Extractor. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Next-generation sequencing (NGS) technologies have accelerated considerably the investigation into the composition of genomes and their functions. Genotyping-by-sequencing (GBS) is a genotyping approach that makes use of NGS to rapidly and economically scan a genome. It has been shown to allow the simultaneous discovery and genotyping of thousands to millions of SNPs across a wide range of species. For most users, the main challenge in GBS is the bioinformatics analysis of the large amount of sequence information derived from sequencing GBS libraries in view of calling alleles at SNP loci. Herein we describe a new GBS bioinformatics pipeline, Fast-GBS, designed to provide highly accurate genotyping, to require modest computing resources and to offer ease of use. RESULTS: Fast-GBS is built upon standard bioinformatics language and file formats, is capable of handling data from different sequencing platforms, is capable of detecting different kinds of variants (SNPs, MNPs, and Indels). To illustrate its performance, we called variants in three collections of samples (soybean, barley, and potato) that cover a range of different genome sizes, levels of genome complexity, and ploidy. Within these small sets of samples, we called 35 k, 32 k and 38 k SNPs for soybean, barley and potato, respectively. To assess genotype accuracy, we compared these GBS-derived SNP genotypes with independent data sets obtained from whole-genome sequencing or SNP arrays. This analysis yielded estimated accuracies of 98.7, 95.2, and 94% for soybean, barley, and potato, respectively. CONCLUSIONS: We conclude that Fast-GBS provides a highly efficient and reliable tool for calling SNPs from GBS data.