Back to the phylogeny view · rendered from sync/2026-10-10/hop_phylogeny/REPORT.md (2026-10-10); figures and tables are in that folder.

Hop pseudo-phylogeny across our full dataset (2026-10-10). TENTATIVE.

What this is: measurements only, no biological interpretation. A hook blocked the phylogeny worker from writing .md files, so the orchestrator wrote this report from the worker's draft.
Where things are:
commands: PIPELINE.sh;
scripts: scripts/;
tables: tables/;
figures: figs/;
raw outputs: /data/hop_data/hop_phylogeny/;
commit: 32c5773.

Q1. What went in?

Libraries: 1,445, selected by the rule in scripts/p02_manifest.py. - Panels in full: - diversity panel: 198; - USDA: 173; - OSU: 455; - Carey 2026: 22; - KR and SW75; - the 13 assembly haplotypes. - Biparental families (7): both parents, plus at most 5 random eligible progeny each (seed 20261010). Havill progeny are drawn only from the 48 confirmed 21058M children. - Not placed: failed QC (≥ 5,000 usable sites, and ≥ 2,000 sites shared with ≥ 50 % of libraries): - 12 OSU libraries; - 8 Zhang libraries; - all 5 Pilgrim DArT progeny (about 4,600 usable sites). - Zhang 2017 HapMap: removed from all drawn trees (PI, 2026-10-10). Its placement-control numbers are kept below.

The drawn tree has 753 tips: - 692 ApeKI-GBS backbone plants; - 61 placed tips: - 30 WGS; - 5 low-pass WGS; - 5 RNA-seq; - 6 GRAS-Di; - 5 PstI+ApeKI GBS; - 13 assembly haplotypes.

Tables: sample_table.tsv (every library), sample_counts.tsv.

Figures:

File	Content
`fig6_highlight_circular.png`	circular tree, highlighted tips only
`fig7_highlight_linear.png`	linear tree, highlighted tips only
`fig8_collapsed_named_clades.png`	clades of up to 60 tips collapsed and named by the founders, wild groups and assemblies they contain
`fig1`, `fig2`	the same tree with three colour rings

Highlighted tips are listed in founders_found_missing.tsv and highlighted_tips.tsv.

Highlighted tips found (assay of the drawn tip): - Cultivated founders and key ancestors: - Brewer's Gold (GBS); - 19043M (GBS); - Fuggle (2 GBS); - Styrian Golding, a Fuggle clone (2 GBS); - Cluster (GBS); - Tettnanger (GBS); - Northern Brewer (2 GBS); - Kent Golding (GBS); - East Kent Golding (GBS); - Hersbrucker (7 GBS); - Hallertauer Mittelfrüh (2 GBS + Carey WGS); - Bullion (GBS + Carey WGS); - Saaz (GBS + the Saaz h1/h2 assemblies); - 21110M (GBS + M21_H1/H2); - Wye Target (GBS), marked as the OB79 carrier. - Wild and feral (coloured by subspecies label):

Group	Tips
European feral	24
neomexicanus-labelled	26
North American wild, unlabelled	53
lupuloides	3
pubescens	3
cordifolius	5

Assemblies: all 13 haplotypes are tips.

Not found in our data: OB79, Late Cluster, Petham Golding, Early Bird.

Q2. How was distance measured?

Sites: the catalogue of 72.0 M SNPs (21110M frame, autosomes; ChrX excluded).
Distance: sum over shared sites of f_i(1−f_j) + f_j(1−f_i), where f is a library's ALT-read fraction. It is divided by the panel haplotype diversity at the same sites. No genotype calls and no GQ filter are used.
Filters: a per-library depth cap; a minimum of 2,000 shared sites per pair.
Missing pairs: filled by nearest-neighbour substitution. The median error is 0.095 on 15,662 pairs hidden for testing.
Checks:
an independent recomputation found a NaN-denominator bug, now fixed (4/4 library pairs pass);
the BIONJ code recovers known test trees exactly.

Q3. A single tree over all assays? It failed.

In the all-libraries tree, the share of plants whose nearest neighbour is the same assay:

Assay group	Same-assay nearest neighbour	Expected
WGS	0.87	0.03
GBS reads	0.95	0.56
Genotype calls	0.94	0.41

That tree is kept only as supplementary figures S1_*.

Q4. The full tree: a GBS backbone with other assays placed on it

Backbone: 692 ApeKI-GBS plants, BIONJ, 100 bootstraps over 5 Mb blocks. - 259 of 689 edges have ≥ 70 % support; - 148 edges have ≥ 95 %.

Placement: affine weighted least squares. - The first rule tried (additive) failed its controls: WGS same-plant 0/3 and same-name 1/6. - Stability: a median of 83 % of bootstraps place a tip on the same edge, and 93 % within two edges.

Controls (same-name pairs are not verified):

Control	Result
Same plant, GBS vs GBS	128/141
Same plant, WGS vs GBS	2/3
Same name, Carey WGS vs panel	6/6
Same name, Zhang vs panel	37/58
Same plant, RNA-seq	0/1
Assembly haplotype vs its own plant	2/6
Zhang progeny vs mother cultivar	17/143
Pedigree edges: parent within child's 10 nearest	7/113 (chance ≈ 1.5)
Chinook ← 63012M	rank 649 (not recovered)
63012M ← Brewer's Gold	rank 34
Celeia ← Fuggle clone	rank 3 (recovered)
Parent vs child across assays (WGS parent vs its own GBS children)	fails (ranks in the hundreds)
Shuffled genotypes	≈ 0 everywhere (fails, as it should)

Where the wild WGS plants attach:

Plants	Attaches to	Bootstrap support
pubescens	OSU 21558m	100 %
CORDI 2/4/7/8	EU_feral_12	65–74 %
KR	EU_feral_23	22 %
MN-767	Neomex_11
lupuloides plants	OSU 21570m / 21595m

Q5. Catalogue-site trees, WGS plants only

Per chromosome (fig3): - pubescens is one clade on 9/9 chromosomes and in 58 of 91 windows of 25 Mb; - cordifolius is one clade on 9/9 chromosomes and in 91/91 windows.

Group quartets (1,000 block bootstraps, all 100 %): - unrooted arrangement: (pubescens, lupuloides) – neomexicanus – (cordifolius, European lupulus); - margins for pubescens' position: 4–5 %; - this split holds in 60 of 91 windows and on 8 of 9 chromosomes. Chr02 is the exception.

Q6. Joint calls without catalogue ascertainment (30 WGS plants)

Data: 1,076,412 SNPs, thinned to one per 2 kb.

Trees (fig5, BIONJ and IQ-TREE ML, GTR+ASC on 377,066 sites): - (lupuloides + neomexicanus) is one clade, 100 % support; - pubescens joins them, forming one American wild clade; - cordifolius sits outside that clade.

Group-mean distances:

Pair	Distance
between the American wild groups	0.13–0.15
cordifolius to the American wild groups	0.27–0.28
American wild groups and cordifolius to European lupulus	0.45

pubescens' exact position: group quartets are near-ties, margins 0.5–1 % (88–96 % of bootstraps).

Ascertainment effect: on the catalogue sites, cordifolius sat nearest European lupulus and pubescens paired with lupuloides. Neither holds in the joint calls, consistent with the catalogue's 13 haplotypes including no pubescens or cordifolius.

Tree agreement (normalised Robinson–Foulds distance): - joint-call vs catalogue BIONJ: 0.33; - BIONJ vs ML: 0.48.

No root is inferred.

Q7. Local affinity along the genome (`fig4`)

Measure: for each plant and 25 Mb window, the nearest WGS reference group.
Coverage: about 540–590 plants per window have enough shared sites with every group.

Caveats

Assay offsets remain.
The placement rule was chosen after the first rule failed its controls.
Catalogue ascertainment: the catalogue comes from 13 haplotypes, none of them pubescens or cordifolius.
Small reference groups: neomexicanus is one plant (MN-767).
Unstable placements: low-pass WGS and GRAS-Di.
No outgroup: no outgroup species is in our data.
Disk: 8.5 GB of RNA CRAMs sit in /data/hop_data/hop_phylogeny/wyerna/cram. Deleting them needs the PI's OK.

Proposal (not built)

A "phylogeny" page on the :8900 atlas: the backbone + placement tree with search, the colour rings, a per-window local-tree slider, and a link from each tip to its atlas row.
Built from: combined.nwk, placements_affine.tsv and local_affinity.tsv.gz.