SEPIA — cluster inventory

A snapshot of everything on the cluster, so the compute side is accounted for in the repo. Not published to the website (this is a .md; the Pages workflow only serves plan/*.html) because it lists internal cluster paths.

Location: /nfs/turbo/lsa-tlasisi1/lheald_thesis/aDNA_data on the University of Michigan Great Lakes HPC system (Turbo storage).
Size: ~110 GB. Snapshot date: 2026-07-22.
Access: SSH to greatlakes (needs a UMich Kerberos password + Duo two-factor; no key-only login). Reuse an authenticated session via a ControlMaster socket, then ssh greatlakes "<cmd>". Scratch working area: /scratch/tlasisi_root/tlasisi0/lheald/. SLURM account tlasisi0.
Reference genome: reference/hs37d5.fa = GRCh37 / hg19 (this is the build all the sequence data is on — see the build-mismatch bug A2 in plan.md). A hg19.fa is referenced by one script but is absent.
To refresh this inventory: ssh greatlakes "du -sh /nfs/turbo/lsa-tlasisi1/lheald_thesis/aDNA_data/*/" etc. Keep cluster commands text-only (ls, du, grep, samtools view -H, bcftools view -H) — never cat/head a .bcf/.bam (binary → scrambles the terminal).

Directory layout (by size)

Directory	Size	Contents
`sgdp/subsets/`	102 GB	326 per-sample SGDP (modern reference) BCF subsets — two SNP-panel subsets per sample. The bulk of the store.
`ancient_wholegenome/`	4.5 GB	Merged archaic genotype calls at the whole-genome SNP panel (`ancient_sgdp_wg.bcf` / `.vcf.gz`).
`pca/`	1.8 GB	`modern/` (SGDP whole-genome PLINK files + PCA eigen-outputs) and `pigmentation/` (pigmentation-panel PLINK + PCA + an `ancient/` subdir with the ancient pigmentation calls).
`reference/`	1.4 GB	`hs37d5.fa` (+ `.fai`) — the GRCh37/hg19 reference.
`snps/`	928 MB	SNP-panel definitions: `snps.bed` (bare-named), `snps.chr.bed` (chr-prefixed), `sgdp.snps.pos`, `sgdp.bim`.
`results/`	5.7 MB	`depth/` — 15 `*_depth.txt` per-sample depth tables (one, `Sid1253`, is empty); `filtered/` — 15 panel-filtered BAMs.
`sgdp_merge/`	112 KB	Alternate SGDP merge scripts (`vcf_wget`, `vcf_merge`, `merge_continue`).
`work/`	64 KB	`bamlist.txt` and scratch bits.
(root)	—	~15 `*.slurm` scripts + `wget_commands.txt` (558 lines = 279 SGDP samples × 2).

The pipeline (~19 SLURM jobs), by stage

Acquire archaic BAMs — sort_wget.slurm, phased_get.slurm (download 15 archaic genomes from MPI-EVA & EBI/ENA; merge per-chromosome, sort, index).
Depth & filtering — snp_subset.slurm (samtools depth -a -b snps.bed → per-sample depth tables; samtools view -L → panel-filtered BAMs).
Ancient genotype calling — ancient_wg.slurm (whole-genome panel, per chromosome; note the --ploidy 1 bug A3); pig_call_miss.slurm (pigmentation panel, keeps all sites — the one used); pigmentation_ancient_call.slurm (pigmentation panel, variants-only — unused; bug A4).
Acquire & subset SGDP — sgdp_subset_array.slurm (279-task array: download each modern VCF, subset to both panels); alternates get_simons.slurm, sgdp_merge/vcf_wget.slurm.
Merge + modern PCA — sgdp_merge_pca.slurm / sgdp_merge_panel.slurm (merge subsets → PLINK → LD-prune → PCA with allele weights); modern_pca.slurm; pig_pca.slurm (pigmentation PCA); subset_sgdp.slurm.
Project archaics onto modern PCA — analysis/pca_proj.py & analysis/pig_pca_proj.py in the repo (plink2 --score); pca_ancient.slurm on the cluster is dead code referencing the missing hg19.fa (bug A5).

Tools: bcftools, samtools, plink/plink2, tabix, eigensoft, wget.

Archaic samples (15 BAMs; 14 analyzed)

Chagyrskaya8, Denisova3, den11 (Denisova 11 / "Denny"), den25 (Denisova 25), den5 (Denisova 5), goyet, hst (Hohlenstein-Stadel), lesCottes, mez1 (Mezmaiskaya 1), mez2 (Mezmaiskaya 2), scladina, Sid1253 (El Sidrón — depth file empty, not analyzed), spy, vindija33 (Vindija 33.19), vindija87 (Vindija 87). Note: Denisova3's BAM uses chr-prefixed contigs while all others (and the reference) are bare-named — see bug A1.

Key derived products (expensive to regenerate → back these up, item C3)

ancient_wholegenome/ancient_sgdp_wg.bcf / .vcf.gz — the merged archaic genotype call set at the whole-genome SNP panel.
sgdp/merged_sgdp_panel.bcf, sgdp/merged_snps_panel.bcf — the merged modern SGDP reference panels.
pca/modern/sgdp.wg.pca.*, pca/pigmentation/pigmentation_snps_pca.* — the PCA eigen-outputs (these are what got copied into the repo's data/).
results/depth/*_depth.txt — the per-sample depth tables (also copied into the repo's data/).

The raw archaic BAMs are re-downloadable from MPI-EVA (ftp.eva.mpg.de/neandertal) and EBI/ENA, so they're lower priority to preserve than the derived products above.