This repository visualizes the phenotypes of H5 HAs calculated as the sum of the effects of the constituent mutations as measured by the pseudovirus deep mutational scanning (DMS) of Dadonaite et al (2024). This visualization workflow was written by Jesse Bloom.
Briefly, the H5 HA protein sequences are extracted from Angie Hinrich's pre-built UShER tree. The protein sequences are then aligned to the protein used in the DMS to identify all mutations relative to that strain, and phenotypes are calculated as the sum of the effects of the constituent mutations in the DMS data.
The data are the provided in several visual forms. Because the extrapolations of the DMS measurements are expected to be most accurate for sequences more closely related to the one used in the DMS, visualizations are made at several different protein identity cutoffs relative to the DMS strain.
These plots group strains by their protein haplotype, and then allow you to plot points representing the different haplotypes. You can mouseover points for details, adjust which DMS phenotypes are shown on the x- and y-axes, filter by divergence from the DMS protein, and scale or hide points based on how many sequences have that haplotypes.
The plots are at:
These trees are the ones built by UShER on the nucleotide sequences, but points are colored by the DMS phenotypes based on the proteins. You can use the dropdown at left to choose which phenotype to use to color the trees, and mouseover points for details:
These plots show how many sequences were retained/filtered for each sequence set shown above:
The file results/usher_prebuilt/version_info.txt indicates which version of the pre-built UShER tree was used.
Here are numerical and sequence data tracked in this repo:
The DMS phenotypes shown are those measured in Dadonaite et al (2024), and are escape from neutralization by ferret or mouse sera, HA stability, and a2-6-linked sialic acid usage.
Phenotypes suffixed by _increase are those that only include mutations with positive effects on the phenotype; this is of relevance as some mutations are measured to decrease (rather than increase) neutralization escape or stability but the biological relevance of those measurements are not always clear.
See the interactive DMS data page for more details on per-mutation effects.
Mutations are shown in three numbering schemes:
See https://dms-vep.org/Flu_H5_American-Wigeon_South-Carolina_2021-H5N1_DMS/numbering.html for more details on sequence numbering.
Note also that UShER itself numbers with respect to A/Goose Guangdong/1/1996 which has an insertion in HA2 relative to the DMS strain.
The sequences used here are extracted from the UShER tree, which can interpolate some missing amino acids during phylogenetic placement. In addition, indels are not included. Therefore, be sure to directly check the identities of any sequences of high interest.
The UShER tree sequence set is not as highly QC-ed as for example the Nextstrain tree to remove erroneous or duplicated sequences, so there may be some spurious or low-quality sequences.
The DMS phenotype of sequences are just calculated as the additive effects of the mutations as measured in DMS. Particularly for sequences with many mutations, this additive model could break down and the phenotypes for mmore highly mutated sequences (relative to the DMS strain) may be less relevant.
The analysis is run by a snakemake pipeline.
The pipeline is in Snakefile, and the configuration is in config.yaml.
The conda envs used by the pipeline are in ./envs/.
To run the pipeline, build the Flu-HA-H5-2.3.4.4-DMS-informed-surveillance environment in envs/global.yml, activate it, and run Snakefile with:
snakemake -j 4 --software-deployment-method condaOn the Hutch cluster, you can just run the pipeline with the bash script run_Hutch_cluster.bash.
Code used by the pipeline is in ./notebooks/ and ./scripts/.
Numerical and sequence results are placed in ./results/, with only some files tracked in the GitHub repo.
The interactive Altair plots are placed in ./docs/, where they can be visualized using GitHub Pages.
The Nextstrain Auspice JSON files are placed in ./auspice/, where they can be visualized as described here.
As the pre-built UShER tree is updated, you will want to re-run the pipeline. To do that, just remove the contents of the ./results/, ./docs/, and ./auspice subidirectories and then just re-run the pipeline.
The script check_for_prebuilt_usher_update.py can be set up to run using crontab -e to check for such updates.