Butterfly

Overview

After initial taxonomic assignment through the Holi pipeline, we merged samples, refined the taxonomic profiles, quantified DNA damage patterns, and developed age-dependent filtering models to distinguish genuine ancient DNA (aDNA) from noise. The workflow integrates filterBAM, metaDMG, and R-based model fitting to assess DNA authenticity and reproducibility across technical and biological replicates.

Taxonomic Refinement

Taxonomic alignments from Holi were post-processed using filterBAM (v1.5.1):

Reassign step:

--iters 0 --min-read-ani 94 --min-read-count 3

Filter step:

--min-read-ani 94 --min-read-count 3 \
--min-expected-breadth-ratio 0.5 --min-normalized-entropy auto \
--min-normalized-gini auto --min-breadth 0 --min-avg-read-ani 90 \
--min-coverage-evenness 0.4 --min-coverage-mean 0 --include-low-detection

These steps ensure high-confidence taxonomic assignments and consistent genome-wide coverage metrics before downstream damage estimation.

DNA Damage Estimation

DNA cytosine deamination and model-based noise were estimated using metaDMG (v0.4-93):

1. LCA assignment — each read's lowest common ancestor (LCA) was determined from all possible alignments.
2. Damage model fitting — for each taxon and replicate, metaDMG estimated parameters describing observed damage (A_b) and noise (c).

Model quality was evaluated using Lin's Concordance Correlation Coefficient (CCC):

Good fit criteria:

• ρ_c ≥ 0.85
• C_b ≥ 0.9
• p-value < 0.1
• Noise baseline confidence interval within [0, 1]

Fits failing any criterion were classified as bad.

DNA Damage–Age Models and Filtering

For each taxon (≥500 reads), we plotted median A_b against sample age:

• DNA damage decreases with younger layers, consistent with expected temporal decay.
• Marine sediments showed lower deamination than contemporaneous lacustrine layers.
• Post-1900 AD samples exhibited little to no measurable damage.

To define genuine aDNA signals, we computed:

• The 5th percentile of A_b across taxa (660 CE–1900 CE)
• Smoothed these thresholds with a LOESS regression
• Retained taxa exceeding the lower 95% CI of the fit

Taxa showing authentic damage in older layers were also accepted in younger ones ("oldest pass date" rule).

Visualization and Reproducibility

• Figures S10–S11: Damage vs. age for individual taxa (e.g., Ovis) and replicate consistency
• Figures S12–S13: PCoA of eukaryotic community profiles across replicates
• Figures S14–S16: Combined DNA damage–age models and taxa-specific trends
• Figures S17–S19: Rarefaction and negative control validation

All plots and summaries were generated in R, integrating metaDMG, filterBAM, and custom scripts for visualization and statistical filtering.

HTML Reports

• SampleWorkflow.html — Overview of pipeline and sample-level workflow
• Replicates.html — Detailed replicate analysis, damage modeling, and filtering results
• Negatives.html — Anaylsis of taxa in blanks