pangolin v4.0

Angie Hinrichs, UCSC Genomics Institute

Theiagen Office Hours

April 12, 2022

Overview

• What’s new in pangolin v4.0
• --analysis-mode: UShER is the default, pangoLEARN uses random forest
• Output columns & version descriptors
• Data repository consolidation
• Optional cache mode for UShER
• Questions!

https://twitter.com/AineToole/status/1509876534529638411

--analysis-mode usher|pangolearn|accurate|fast|scorpio

• UShER is default
• pangoLEARN is still available via --analysis-mode pangolearn
• Aliases:
• --analysis-mode fast = pangolearn
• --analysis-mode accurate = usher

And in pangolin v4.0.3:

• --analysis-mode scorpio

PangoLEARN model change

• v3 pangoLEARN model: decision tree
• v4 pangoLEARN model: random forest of decision trees
• Hopefully more robust
• Requires more RAM (>= 16GB or 32GB or … depending on how many sequences)

Changes to output columns

pangolin v3 pangolin v4�---------------------------------------------�taxon taxon�lineage lineage�conflict conflict�ambiguity_score ambiguity_score�scorpio_call scorpio_call�scorpio_support scorpio_support�scorpio_conflict scorpio_conflict� > scorpio_notes scorpio call: Alt alleles 21; Ref alleles…�version version (PUSHER|PLEARN|PANGO)-v1.3�pangolin_version pangolin_version 4.0.4�pangoLEARN_version <�pango_version <� > scorpio_version 0.3.16� > constellation_version v0.1.4 �status <� > is_designated True|False� > qc_status pass|fail� > qc_notes Ambiguous_content:0.02 �note note Usher placements: BA.1(1/1); scorpio replaced…

Version descriptors

In pangolin v3:

• pangoLEARN_version = date on which pangoLEARN training started
• pango_version = pango-designation release used to train pangoLEARN

In pangolin v4:

• No more date, just pango-designation release in version e.g. (PUSHER|PLEARN|PANGO)-v1.3

Data dependencies (github repositories)

In pangolin v3:

• pangoLEARN: Trained pangoLEARN model and UShER tree
• pango-designation: Alias file (e.g. AY = B.1.617.2, BA = B.1.1.529, …)
• Confusing because pangoLEARN also depends on pango-designation and often lags it by weeks

In pangolin v4:

• pangolin-data: Trained pangoLEARN model, UShER tree, alias file

Optional cache mode for UShER assignments

Usher is slower, but UCSC precomputes UShER assignments on all GISAID sequences.

Replace GISAID sequence names with sequence hash value → lookup table.

Big file (~200MB) – not included in pangolin by default.

To install the cache:

pangolin --add-assignment-cache

To use the cache:

pangolin --use-assignment-cache

Resources

pangolin documentation:�https://cov-lineages.org/resources/pangolin.html

pangolin bug reports/questions:�https://github.com/cov-lineages/pangolin/issues/

New Pango lineage requests:�https://github.com/cov-lineages/pango-designation/issues

“Designated” vs. “Assigned” lineages:�https://www.pango.network/how-does-the-system-work/genome-designation-versus-assignation/

Pango lineage rules:�https://www.pango.network/the-pango-nomenclature-system/statement-of-nomenclature-rules/

Acknowledgements

• U. of Edinburgh: Àine O’Toole, Emily Scher, Rachel Colquhoun, Andrew Rambaut (pangolin)

​

• UCSD: Yatish Turakhia, Cheng Ye (UShER, matOptimize)

​

• UCSC: Russ Corbett-Detig, Jakob McBroome, Bryan Thorlow, Adriano de Bernardi Schneider, Alex Kramer, Marc Perry (matUtils, evaluation)

What defines a Pango lineage?

Not a set of mutations!

lineages.csv in the pango-designation github repository (>1M lines):

​

...

India/GJ-ICMR-NIV-INSACOG-GSEQ-3045/2021,B.1.617.2

India/PY-SEQ_294_S22_R1_001/2021,B.1.617.2

Malaysia/IMR_682164/2021,B.1.617.2

Japan/IC-1175/2021,B.1.617.2

USA/TX-CDC-ASC210037740/2021,B.1.617.2

England/WSFT-25C6539/2021,B.1.1.7

USA/MI-UM-10039543606/2021,AY.3

USA/KS-KHEL-1922/2021,AY.3

USA/KS-KHEL-1923/2021,AY.3

USA/MO-MSPHL-002099/2021,AY.3

USA/MO-MSPHL-002132/2021,AY.3

...

Use UShER!

Some incongruous changes with UShER…

Many incongruous changes with pangoLEARN

When new lineages are added…