Bentley et al. estimated that 15X mapped read depth of WGS samples would be sufficient to detect almost all homozygous SNPs and 33X for almost all heterozygous SNPs [12]. 50X was estimated by Ajay et al. for all SNPs and small indels [13]. Depending on the capture kit, Clark et al. calculated that exome-seq required 80X mean on-target depth to reach the common threshold of 10X per-site depth in 90% or more of all targeted regions [10].
In general, we recommend GATK-HC for variant calling and filtering for several reasons. First of all, GATK-HC outperformed SAMtools-mpileup in most of our situation tests resulting a higher precision and recall ratio for SNP and INDEL detection. Second, GATK-HC allows rapid incorporation of multiple samples into a dataset without needing to recall genotypes for all samples, even previously genotyped
ones, from aligned reads by using the GVCF system. This saves considerable time and computational expense when adding samples to a dataset. The third reason to recommend GATK-HC is that it supports multi-thread processing which is not available in the SAMtools-mpileup. Taking the advantage of high-performance clusters, multi-thread feature can significantly save processing time especially for large studies. Finally, the GATK package supports sophisticated machine learning based variant filtering (VQSR) which
showed superior performance than empirical hard cutoffs. We did, however, find situations that SAMtools-mpileup is more preferable depending on the goal of the study. For example, for a low diversity population with very low sequencing coverage (1x), SAMtools-mpileup was able to identify more true SNPs than GATK-HC but at the cost of lower precision. If the purpose of the experiment is to identify as many true positive SNPs as possible, then SAMtoolsmpileup could be used in this particular situation. Another situation that SAMtools-mpileup may be preferable is identifying SNPs from a closely related sample. According to the simulation results from single samples, SAMtools-mpileup resulted slightly higher precision and recall values than GATK-HC results when the mutation rate was lower than 0.05. If the experiment aims at charactering SNPs in a genotype that is closely related to the reference genome, SAMtools-mpileup could be used in this particular situation.