Zoonotic risk assessment from viral genomes

Code and data used in Mollentze N, Babayan SA, Streicker DG (2021) Identifying and prioritizing potential human-infecting viruses from their genome sequences. PLOS Biology 19(9): e3001390. doi:10.1371/journal.pbio.3001390

For a list of priority categories and ranks for all virus species in the paper, see here.

Note: This repository uses git-lfs. This must be installed before cloning to ensure large files are included.

Table of contents

• Requirements
• Ranking novel viruses
• Repeating all analyses
• File structure

Requirements

• Install the conda package manager
• Create the base environment (this installs everything required for prediction of new viruses)

conda env create -f base_environment.yml

• Before each use, activate this environment using

conda activate zoonotic_rank

Repeating published analyses

If repeating all analyses in the manuscript (see below), a few additional tools and R libraries are needed. The majority of these can be added to the base environment created above using:

conda env update -n zoonotic_rank -f dev_environment.yml

The BLAST+ suite of applications is also required (used for "phylogenetic neighbourhood" analyses and predictions). Version 2.8.1+ was used in the manuscript. If your R session has trouble finding the BLAST+ executables, run make update_path and enter the location of the BLAST executables (e.g. /usr/local/ncbi/blast/bin).

Ranking novel viruses

Ranks for novel viruses can be generated by specifying the input sequence format, paths to genome and metadata files, and a name for output files, e.g:

Rscript Scripts/PredictNovel.R fasta \
                               InternalData/example_files/genomes.fasta \
                               InternalData/example_files/metadata.csv \
                               example_1

For detailed instructions and further options, see

Rscript Scripts/PredictNovel.R --help

Input

The example files in InternalData/example_files/ illustrate the required file format for the most complicated scenario, in which sequences are in fasta format and open reading frames are specified manually. In such cases, the metadata file should have the following form (excluding the "Comments" column):

Name	SequenceID	CodingStart	CodingStop	Comments
Non-segmented virus 1	NC_031751.1	107	1894	Each line specfies an open reading frame (ORF)
Non-segmented virus 1	NC_031751.1	2309	4252	Name and SequenceID repeated to specify another ORF in the same sequence
Segmented virus 1	NC_015451.1	20	6277	Multiple viruses are distinguished using the Name column
Segmented virus 1	NC_015450.1	24	3986	Name repeated to specify a further SequenceID belonging to this virus
Segmented virus 1	NC_015452.1	36	866	The first of two genes on this segment
Segmented virus 1	NC_015452.1	1778	1041	Coordinates reversed to indicate that the ORF is on the complementary strand

If sequences are available in genbank format, only the first two columns are required, mapping each virus name to one or more sequence identifiers. This format allows more flexiblility in the specification of open reading frames (see the "FEATURES" section of https://www.ncbi.nlm.nih.gov/genbank/samplerecord).

Output

All files generated will be preceded by the output name specified ("example_1" in the example above).

File	Description
*.predictions.csv	Predicted scores, binary predictions of whether the virus is expected to infect humans, and priority categories.
*.genome_features.csv	Calculated genome composition features used to make predictions.
*.explanations.csv	Influence of individual features on the predicted scores of each virus (i.e., SHAP values).
*.similar_explanations.csv	Viruses in the training data which have similar explanations for their predicted scores. All viruses are clustered based on similarity in their SHAP values, using affinity propagation clustering to identify discrete clusters and the most central observation in each cluster (the exemplar). This output file details clusters containing at least one of the input viruses.
*.all_cluster_members.csv	All clusters obtained during the above SHAP similarity-based clustering.
*.similar_explanations.pdf	Dendrogram illustrating the distances between clusters in *.all_cluster_members.csv. Red branches highlight clusters containing input viruses.
*.feature_dendrogram.pdf	Hierarchical clustering of viruses by similarity in their genome features, giving an indication of how distant the input viruses are from the training data. Note that this clustering will not neccesarily reflect taxonomy.

Repeating all analyses

Follow instructions below to repeat the analyses described in the manuscript. Note that these steps are not needed to make predictions as described above (pre-trained models are included). Running all analyses takes ~3 weeks on a 4-core, 2.8 GHz i7 processor and requires ~5Gb of disk space.

Basic

These steps will download all external data and re-run the entire pipeline.

Using Rstudio:

1. Open zoonotic_rank.Rproj in RStudio
2. On the Build tab, select More > Clean and Rebuild

Using the command-line:

make clean all

Advanced options (command-line only)

• Use make help to see individual steps in the pipeline. Upstream steps are run automatically if needed. For example, using make prepare will run the data cleanup step, but also downloads the raw data if needed.
• make as_distributed resets the project to the state in which it was distributed.
• make clean removes all run-related files, allowing a complete re-run (in contrast to as_distributed, this includes removing the pre-trained models required for prediction).

File structure

(files/folders which will be created during a full run are indicated by [*])

└─zoonotic_rank/
   ├─Makefile ................................. Record of workflow and dependencies between files
   ├─options.config ........................... Runtime options (specifies number of parrallel
   │                                            threads allowed and the random seed)
   ├─base_environment.yml ..................... Record of software and R libraries required for 
   │                                            prediction
   ├─dev_environment.yml ...................... Record of additional software required to train and
   │                                            evaluate models
   ├─FigureData/ ...............................Underlying values for manuscript figures
   ├─InternalData/ ............................ All data unique to this project
   │   ├─example_files/ ....................... Example input files for predicting novel viruses
   │   ├─Shaw2017_raw/ ........................ Raw ISG data from Shaw et al. 2017
   │   │                                        (see https://isg.data.cvr.ac.uk/)
   │   ├─FinalData_Cleaned.csv .................Final dataset, as used for training. Created by 
   │   │                                        merging files below (Scripts/MergeAndCleanData.R)
   │   ├─AllInternalData_Checked.csv .......... Metadata for the viruses used as training data
   │   ├─Final_Accessions_Unique_Spp.csv ...... Accession numbers of sequences used for training
   │   │                                        (replaces those in the metadata file)
   │   ├─NameMatches_All.csv .................. Manually curated list used to match virus names to
   │   │                                        unique species across external datasets
   │   ├─SourcesOfZoonoses_BabayanZoonotic.csv  Additional zoonotic status data for species not
   │   │                                        available in external data sources
   │   └─Taxonomy_UnclassifiedViruses.csv ..... Taxonomic information for unclassified viruses in
   │                                            the metadata (unused)
   │
   ├─CalculatedData/ .......................... Intermediate calculations ([*], except for files
   │                                            required by PredictNovel.R)
   ├─ExternalData/ ............................ [*] Data from external sources, dowloaded as needed
   │                                            (see Makefile)
   ├─Misc/ .................................... Miscelaneous scripts to download external data
   ├─Plots/ ................................... [*] Final plots generated
   ├─Predictions/ ............................. [*] Predictions for case studies
   ├─RunData/ ................................. Trained models ([*], except for files required 
   │                                            by PredictNovel.R)
   ├─Scripts/ ................................. Main analysis, prediction, and plotting scripts
   │   └─Plotting/ ............................ Scripts to generate published plots
   ├─Tests/ ................................... Unit tests for basic functionality of utility 
   │                                            functions/scripts
   └─Utils/ ................................... Utility functions and tools called by other scripts