Skip to contents

bioclients: Clients for Biological Database Web Services

Source: R/bioclients-package.R
bioclients-package.Rd

Look up genes, variants and proteins from R, without writing a client for every biological web service. Each service gets one client that makes the request and returns a table. Parsing is a separate function that needs no network, so it can run on a saved response and be tested offline. Transport, retries, caching and error handling are left to the 'biohttp' package. Dependencies for single services are optional, so you do not install what you will not use. The services covered include 'Ensembl', described in Dyer et al. (2025) doi:10.1093/nar/gkae1071 , 'UniProt', in The UniProt Consortium (2025) doi:10.1093/nar/gkae1010 , 'gnomAD', in Chen et al. (2024) doi:10.1038/s41586-023-06045-0 , 'Open Targets', in Buniello et al. (2025) doi:10.1093/nar/gkae1128 , and the 'AlphaFold' Protein Structure Database, in Varadi et al. (2024) doi:10.1093/nar/gkad1011 . Each client's help page cites the service it calls.

Details

Every service module ships two halves, and the split is the point.

The client builds a request, calls into biohttp, and returns its envelope. It knows URLs, parameters, and rate limits, and it touches the network. Start at mygene_gene(), gnomad_constraint(), gnomad_frequency(), or clinvar_classification().

The parser takes an already-parsed response body and returns a canonical structure. It is pure, it never touches the network, and it is what the offline tests exercise. See mygene_parse_hits(), gnomad_parse_constraint(), and clinvar_parse_record().

A caller that already has a response body can use a parser on its own.

What comes back

Every client returns a biohttp envelope rather than raising, so a caller branches on res$status and writes no tryCatch() of its own. Reach for biohttp::body_or_null() when the reason for a failure is genuinely not actionable.

Reaching a source that has nothing for a query is no_data, not an error. It is an answer.

Batching

mygene_genes() and gnomad_constraints() ask about many genes in one or a few requests rather than one request per gene. The round trip is where essentially all the time in these calls goes, so this is the difference that matters. Both return rows in input order, one row per input, so a caller can zip results onto its inputs by position.

See also

Useful links:

Author

Maintainer: Samuel Bharti samuelbharti.io@gmail.com (ORCID) [copyright holder]

Authors:

Other contributors: