SARS-CoV-2 lineage assignments using phylogenetic placement/UShER are superior to pangoLEARN machine-learning method

Adriano de Bernardi Schneider; Michelle Su; Angie S Hinrichs; Jade Wang; Helly Amin; John Bell; Debra A Wadford; Áine O'Toole; Emily Scher; Marc D Perry; Yatish Turakhia; Nicola De Maio; Scott Hughes; Russ Corbett-Detig

doi:10.1093/ve/vead085

SARS-CoV-2 lineage assignments using phylogenetic placement/UShER are superior to pangoLEARN machine-learning method

Virus Evol. 2024 Jan 11;10(1):vead085. doi: 10.1093/ve/vead085. eCollection 2024.

Authors

Adriano de Bernardi Schneider^{1

2}, Michelle Su³, Angie S Hinrichs¹, Jade Wang³, Helly Amin³, John Bell⁴, Debra A Wadford⁴, Áine O'Toole⁵, Emily Scher⁵, Marc D Perry¹, Yatish Turakhia⁶, Nicola De Maio⁷, Scott Hughes³, Russ Corbett-Detig^{1

2}

Affiliations

¹ Genomics Institute, University of California Santa Cruz, Santa Cruz, CA 95064, USA.
² Department of Biomolecular Engineering, University of California Santa Cruz, Santa Cruz, CA 95064, USA.
³ Department of Health and Mental Hygiene, New York City Public Health Laboratory, New York, NY 10016, USA.
⁴ California Department of Public Health (CDPH), VRDL/COVIDNet, Richmond, CA 94804, USA.
⁵ Institute of Evolutionary Biology, University of Edinburgh, Edinburgh EH9 3FL, UK.
⁶ Department of Electrical and Computer Engineering, University of California San Diego, San Diego, CA 92093, USA.
⁷ European Molecular Biology Laboratory, European Bioinformatics Institute, Hinxton CB10 1SD, UK.

Abstract

With the rapid spread and evolution of SARS-CoV-2, the ability to monitor its transmission and distinguish among viral lineages is critical for pandemic response efforts. The most commonly used software for the lineage assignment of newly isolated SARS-CoV-2 genomes is pangolin, which offers two methods of assignment, pangoLEARN and pUShER. PangoLEARN rapidly assigns lineages using a machine-learning algorithm, while pUShER performs a phylogenetic placement to identify the lineage corresponding to a newly sequenced genome. In a preliminary study, we observed that pangoLEARN (decision tree model), while substantially faster than pUShER, offered less consistency across different versions of pangolin v3. Here, we expand upon this analysis to include v3 and v4 of pangolin, which moved the default algorithm for lineage assignment from pangoLEARN in v3 to pUShER in v4, and perform a thorough analysis confirming that pUShER is not only more stable across versions but also more accurate. Our findings suggest that future lineage assignment algorithms for various pathogens should consider the value of phylogenetic placement.

Keywords: Bioinformatics; COVID-19; Phylogenetics; variants.