All structures registered in 3decision are automatically classified against a curated set of protein families as part of the post-registration analysis pipeline. Family definitions are based on the protein family definitions used in MOE and provided by Chemical Computing Group (CCG).
Each protein family is represented by a high-quality multiple sequence alignment containing carefully selected reference sequences and structures. From this alignment, a Hidden Markov Model (HMM) is generated to capture the characteristic sequence signature of the family.
Protein family classification is performed at the biomolecular sequence level rather than directly on individual structures. Biomolecular sequences in 3decision, including UniProt sequences and generic PRO_ sequences, are compared against the family HMMs using HMMER. A sequence is assigned to a family when its match score falls below the family-specific E-value cutoff.
When a structure is registered, its chains are mapped to biomolecular sequences. Protein family matches are then projected from the biomolecular sequence onto the corresponding residue range represented in the structure. As a result:
If a newly registered structure contains a sequence that is not yet represented in 3decision, a new PRO_ biomolecular sequence is created. This sequence is then searched against all supported family HMMs using HMMER to determine whether it belongs to one or more protein families and to assign any applicable family annotations.
The Family definitions also include biologically meaningful annotations that identify important regions within family members. Examples include:
These annotations are defined on the family's reference alignment and projected onto matching biomolecular sequences based on their alignment to the family HMM. Through residue-level mapping, the corresponding annotation ranges become available on structure chains wherever they overlap the resolved portion of the sequence.
In the Antibody family, the CDR annotation sets are CDR_L1, CDR_L2 and CDR_L3 on light chains, and CDR_H1, CDR_H2 and CDR_H3 on heavy chains. The CDR positions were defined for each sequence of the family dataset with the CCG numbering of MOE, which implements CDR definitions based on the Martin numbering scheme.
References:
To show the CDRs of a structure, see How to show the CDRs of an antibody.
The current integration supports a fixed set of 24 protein families, covering major drug discovery target classes and therapeutic modalities.
Once a structure has been classified, family assignments and associated annotations become available throughout 3decision.
The Protein Families section of the Information Browser displays:
The Annotation Browser has a Protein Family track for each family, with:
Family information can be visualised directly in the structure viewer:
Protein Family can be used as a search criterion when querying the database.
Because 3decision also supports the existing ChEMBL-based protein classification, family options are labelled according to their source:
Protein Family can be used as a filter to focus results on a specific target class or therapeutic modality.
Protein family classification and annotation for an existing structure are refreshed if the Sequence Mapping analysis is relaunched. Once the sequence mapping has been updated, 3decision automatically reassesses any applicable protein family assignments and annotations.