ATLAS: Analysis Tools for Low-depth and Ancient Samples

Key results

1. SuggestedIn simulated ancient single-end data with PMD, the dedicated ATLAS pipeline made increasingly fewer genotyping errors than GATK combined with mapDamage 2.0 at higher sequencing depths, though not at low depth.

   One simulated diploid chromosome of ten Mbp (θ = 0.01) with sequencing errors and PMD at commonly observed rates; both tools run with comparable parameters (BQSR recalibration, maximum-likelihood calls), GATK with mapDamage 2.0 damage correction. No error-rate values are given in the text. · section 1; 3 Results

2. SuggestedATLAS calls showed no reference bias, whereas GATK's errors and missing calls concentrated at homozygous-alternative sites; the authors argue this gives ATLAS more power to detect real differences between ancient and modern samples.

   GATK made about ten times fewer genotyping errors at true Ref/Ref than at Alt/Alt sites, depending on depth, and emitted fewer calls at Alt/Alt sites, especially at low depth; alternative GATK parameterisations did not remove this (supplementary section 4). ATLAS showed no such bias in the same simulation. · 3 Results

3. SuggestedAt low depth ATLAS had slightly higher error rates, because it emits a low-quality call at every site with data while GATK skips many such sites, but it called a larger fraction of the genome correctly.

   Same simulation; the text attributes the difference to how each tool handles uncertain genotypes (ATLAS calls them with low quality scores, GATK leaves many uncalled). Error rates and genome fractions are not given as numbers in the text. · 3 Results

4. ClaimedATLAS packages an aDNA/low-depth workflow from raw BAM files: PMD pattern inference, two base-quality recalibration methods, three genotype callers, and PMD-aware tools for heterozygosity, contamination, PSMC and BEAGLE input.

   Described as features; only BQSR recalibration with maximum-likelihood calls is benchmarked here. PMD: position-specific or exponential-decay model. Recalibration: BQSR extension needing known polymorphisms, or reference-free using haploid or ultra-conserved regions. Callers: maximum likelihood, Bayesian with a local-diversity prior, Bayesian haploid caller for very low depth; gVCF output for joint GATK calling. · Step 2: Inferring PMD patterns; Step 3: Recalibrating base quality scores; Step 4: Variant Calling; 2.2 Additional functionalities