Target Questions
Last updated
Last updated
More details can be found in this Blog post. Target predictions can be found on the interface and predictions can be run using this Docker image.
We annotate binding site information in two different ways:
1) For drugs and clinical candidates, where the target is a protein complex and we know which subunit(s) the drug binds to, we indicate this by including a site_id in the drug_mechanism table. This information can be viewed on the ChEMBL interface through the Drug Mechanisms view where the data is found in the Binding_site_name and Binding_site_comment fields.
2) For assay/bioactivity data, we have an algorithm that attempts to predict the most likely binding-domain of the protein target (based on the domain structure of the protein and a list of all curated known small molecule binding domains). This algorithm is only applied to a subset of the > 15 million bioactivity records - binding or functional assays with a single protein or protein complex targets and dose-response measurements - and is not able to make an unambiguous prediction in all cases. There are around 650 K activity measurements with a predicted domain in the predicted_binding_domains table.
Pfam-A domains can be found in the domains table and can be viewed in the Domain Cross References section of the target report card on the ChEMBL interface.
The relationship_type flag provides the relation between the reported target in the source document and the assigned target. D Direct protein target assigned H Homologous protein target assigned M Molecular target other than protein assigned N Non-molecular target assigned S Subcellular target assigned U Default value - Target has yet to be curated
UniProt accessions are the main identifier for protein targets. These can be mapped to Entrez IDs using the UniProt mapping tool. ChEMBL protein targets may be a single protein (with a single accession) or a protein complex, protein family (with more than one associated accession).
It’s possible to search ChEMBL for similar proteins using the EBI-wide BLAST search tool through the ChEMBL interface.
On the ChEMBL homepage, click on 'Advanced Search’ and select ‘Biological Sequence’ and then input your target protein. Similar proteins will be returned alongside the E-value.
We don’t currently have an API endpoint for the protein similarity searches but all target protein sequences are provided in the component_sequences table and can be extracted for further analyses.
This is based on both the target_type and the component_type (to deal with RNA targets) and is shown below.
We map proteins to UniProt accessions and parent and variant proteins will have the same target ChEMBL_ID. Variants are also mapped to a variant_ID that distinguishes these from the wild-type protein (where the variant_ID is null).
Variants may contain single or multiple mutations which include defined (e.g. single amino acid changes) or undefined (e.g. ‘deletion mutants’, mapped to variant_ID -1) changes. The variant sequences table contains details of mutated residues and the sequence.
Variant information is also available through the ChEMBL interface and can be viewed when browsing assays by adding additional columns (variant_sequence_accession and variant_sequence_mutation) and though web services (https://www.ebi.ac.uk/chembl/api/data/assay.json?variant_sequence__isnull=false). The variant_ID can be used to exclude or include protein variants when extracting bioactivity data.
Mutant targets may be associated with with drug resistance or disease but may also include some engineered mutations (users need to review the assay description or references to interpret the data).
The variant_sequences table is not linked to the component sequences table. It’s possible that two assays, one describing a variant and one describing a wild-type protein, may refer to the same protein target (and share a ChEMBL target ID), but provide different accessions. We keep both accessions in case a slightly different sequence was used as the reference to generate the variant numbering in the corresponding assay.
In ChEMBL, we map isoforms to the primary UniProt accession. Sometimes researchers test different isoforms in separate experiments and in these cases one assay is recorded per isoform. The isoform details are recorded in the assay descriptions.
For example:
Assay CHEMBL672502 Binding affinity against Dopamine receptor D2L
Assay CHEMBL670556 Binding Affinity of Compound of Dopamine receptor D2S
Although ‘fuzzy’ matching is available though the UI, unfortunately, we don’t have any fuzzy API support at the moment. However, there is the option to use regular expressions -
e.g. https://www.ebi.ac.uk/chembl/api/data/target?pref_name__iregex=cdk1|cdk2
We extract bioactivity data from a selected set of journals, patents and deposited data sets and include a range of assays such as cytotoxicity assays, antibacterial assays and protein inhibition assays. The extracted data is curated and mapped to ChEMBL targets where possible. However, the curation is an ongoing process and some assays may not be mapped to a target and appear as ‘Unchecked’. There are a number of reasons why the target of an assay may be ‘Unchecked’. For example, the assay may be a physiochemical assay (such as solubility determination) where there is no target, a selectivity ratio where there isn’t a single target and/or the target may be ambiguous, has not yet been created or has not yet been reviewed.
ChEMBL targets are all associated with a unique target ChEMBL_ID. We use UniProt accessions as our primary identifier for protein targets. However, gene symbols can be obtained from the component_synonyms table:
This webservices example may also be useful - https://chembl.gitbook.io/chembl-interface-documentation/web-services/chembl-data-web-services#having-a-list-of-molecules-chembl-ids-in-a-csv-file-produce-another-csv-file-that-maps-every-compound-id-into-a-list-of-human-gene-names.
