BioLiP2 Ligand Relevance and Binding Residues on PDB Structures
View a ligand-bound PDB structure in 3D, mark the residues that touch a named ligand, and judge whether the ligand is biologically relevant.
Get This Design
Clone or download from the repository
Quick start:
git clone https://github.com/davidjurgens/potato-showcase.git
cd potato-showcase/science/genomics/biolip2-ligand-binding-site-review
potato start config.yamlAbout this dataset
BioLiP2 is a database of biologically relevant ligand-protein interactions extracted from the Protein Data Bank, built by Zhang et al. at the University of Michigan and described in Nucleic Acids Research in 2023. The paper notes that many structures are determined with purification and crystallization additives unrelated to a protein's function, so picking out the biologically relevant interactions is nontrivial. At the time of writing, BioLiP2 held 385,160 protein chains involved in 781,684 protein-ligand interactions.
BioLiP2 makes the call with rules and literature checks, not an annotation crowd. A residue with at least two atomic contacts to a ligand is a binding residue, with contacts defined as atom pairs within the sum of van der Waals radii plus 0.5 angstroms. A ligand on a list of 463 commonly used non-biological ligands is discarded if the ligand appears more than 15 times in the structure or the ligand's site holds only two consecutive residues, and the remaining candidates are kept only when the ligand's name appears in the title or abstract of the structure's primary citation. The design asks a person to make the same judgment by eye. Each item names one ligand in a public PDB entry. The annotator selects binding residues in the 3D view or the residue table with a viewer_selection scheme, then answers a radio question with three choices, biologically relevant, crystallization or purification additive, or cannot tell. The cannot-tell choice is an addition of the design. The labels are annotator judgments and not BioLiP2 labels.
No BioLiP2 labels are copied, as a design choice, and items carry no reference answers. The paper's data availability statement says the database and source code are available under the BSD license. The ten structures are PDB archive files, which the wwPDB releases under CC0, and the browser loads the files directly from files.rcsb.org. The design has no model_training block. No Potato trainer reads a protein structure or a viewer_selection scheme, and radio trainers learn from an item's text or image, which here is only a one-line task note. The structure viewer needs WebGL, and the residue table works without WebGL.
- Protein chains in BioLiP2
- 385,160
- Protein-ligand interactions in BioLiP2
- 781,684
- Commonly used non-biological ligands on the artifact list
- 463
- Binding residue rule
- At least two atomic contacts with the ligand
- Sample items
- 10 CC0 PDB entries loaded from RCSB, no BioLiP2 labels
Configuration File
This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.
# BioLiP2: judging ligand relevance and marking binding residues on PDB structures
#
# Dataset: BioLiP2, a database of biologically relevant ligand-protein interactions
# extracted from the Protein Data Bank (PDB).
# URL: https://zhanggroup.org/BioLiP/
# Paper: Zhang, C., Zhang, X., Freddolino, L., and Zhang, Y. (2023). BioLiP2: an updated
# structure database for biologically relevant ligand-protein interactions. Nucleic Acids
# Research 52(D1), D404-D412.
# @article{Zhang_2023, title={BioLiP2: an updated structure database for biologically
# relevant ligand-protein interactions}, volume={52}, DOI={10.1093/nar/gkad630},
# number={D1}, journal={Nucleic Acids Research}, publisher={Oxford University Press (OUP)},
# author={Zhang, Chengxin and Zhang, Xi and Freddolino, Lydia and Zhang, Yang},
# year={2023}, pages={D404-D412}}
# DOI: 10.1093/nar/gkad630
# Earlier version: Yang, J., Roy, A., and Zhang, Y. (2012), Nucleic Acids Research 41(D1),
# D1096-D1103, DOI 10.1093/nar/gks966.
#
# Licences: the BioLiP2 labels are not redistributed here, as a design choice. The paper's
# Data availability statement says the database and source code are available under the
# BSD license. The structures are PDB archive files, which the wwPDB
# releases under CC0 1.0 (https://www.rcsb.org/pages/policies). The ten structures are not
# copied into this folder. The browser fetches each one from files.rcsb.org (allow_remote).
#
# Protocol quotes (Zhang et al. 2023, Nucleic Acids Research):
# "Since many structures are determined with purification and crystallization additives that
# are unrelated to a protein's in vivo function, it is nontrivial to identify the subset of
# protein-ligand interactions that are biologically relevant."
# "A protein residue with at least two inter-molecular atomic contacts to a ligand is
# labeled as a ligand binding residue." Contacts are "protein-ligand atom pairs within sum
# of van der Waals radii plus 0.5 A".
# "A ligand on the artifact list will be discarded if it appears >15 times in the structure,
# or if its binding site only contains two consecutive residues. Otherwise, the title and
# abstract of the primary citation describing the structure is downloaded from PubMed to
# check the remaining potential artifacts."
#
# Adaptations:
# - BioLiP2 decides relevance with geometric rules, a list of 463 commonly used
# non-biological ligands, and a check of the primary citation's title and abstract. It
# uses no annotator crowd. This design asks a person to make the same call by eye, with the
# structure in front of them, so the labels here are annotator judgments and not BioLiP2
# labels.
# - No BioLiP2 annotation is copied. Items name one ligand in a public PDB entry and carry no
# reference answer.
# - The third radio label, "Cannot tell from the structure", is an addition of this design
# and has no source in BioLiP2 or its paper.
# - The radio labels paraphrase the paper's wording ("biologically relevant" versus
# additives for "purification and crystallization"). They are not an official label set.
# - Binding residues are picked by hand in the 3D view or the residue table. BioLiP2 computes
# them from the contact rule quoted above. The rule is given in the instructions as a guide
# and is not applied by Potato.
#
# Model training: none. No Potato trainer reads a protein structure or a viewer_selection
# scheme. Trainers for radio schemes learn from an item's text or image, and the only text
# here is the one-line task note, so a classifier would learn nothing about ligands.
#
# Run (the structure viewer needs WebGL in the browser and network access to files.rcsb.org;
# without WebGL the residue table still works):
# potato start config.yaml -p 8000
annotation_task_name: "BioLiP2 ligand binding-site review"
task_dir: "."
data_files:
- "sample-data.json"
item_properties:
id_key: "id"
text_key: "note"
output_annotation_dir: "annotation_output"
export_annotation_format: jsonl
user_config:
allow_all_users: true
users: []
require_password: false
annotation_instructions: |
Each item names one ligand in a public PDB structure. Find that ligand in the 3D view, then
decide whether it is a biologically relevant partner of the protein or an additive used for
purification or crystallization. Select the protein residues that touch the ligand. BioLiP2
counts a residue as binding when it has at least two atomic contacts with the ligand, with
contact meaning a distance within the sum of the van der Waals radii plus 0.5 angstroms.
Judge by eye and select the residues you think meet that description.
instance_display:
fields:
- key: note
type: text
label: "Ligand to review"
- key: structure
type: protein_structure
label: "PDB structure"
display_options:
engine: 3dmol
allow_remote: true
format: pdb
color_by: chain
show_ligands: true
residue_table: true
height: 460
annotation_schemes:
- annotation_type: viewer_selection
name: binding_residues
description: "Select the protein residues that form the binding site of the named ligand"
viewer_field: structure
labels:
- "binding residue"
min_selections: 1
label_requirement:
required: true
instructions: >-
Click residues in the 3D view or in the residue table. Click a selected residue again to
remove it.
- annotation_type: radio
name: ligand_relevance
description: "Is the named ligand biologically relevant to this protein, or a crystallization or purification additive?"
labels:
- name: "Biologically relevant"
key_value: "1"
- name: "Crystallization or purification additive"
key_value: "2"
- name: "Cannot tell from the structure"
key_value: "3"
label_requirement:
required: true
Sample Data
[
{
"id": "biolip2_001",
"pdb_id": "1STP",
"ligand": "BTN",
"structure": "https://files.rcsb.org/download/1STP.pdb",
"note": "PDB 1STP (streptavidin). Review ligand BTN, chain A, residue 300."
},
{
"id": "biolip2_002",
"pdb_id": "3PTB",
"ligand": "BEN",
"structure": "https://files.rcsb.org/download/3PTB.pdb",
"note": "PDB 3PTB (beta-trypsin). Review ligand BEN, chain A, residue 1."
}
]
// ... and 8 more itemsGet This Design
Clone or download from the repository
Quick start:
git clone https://github.com/davidjurgens/potato-showcase.git
cd potato-showcase/science/genomics/biolip2-ligand-binding-site-review
potato start config.yamlDataset & paper
Zhang et al., Nucleic Acids Research 2023
- Paper DOI
- 10.1093/nar/gkad630
Citation (BibTeX)
@article{Zhang_2023,
title = {BioLiP2: an updated structure database for biologically relevant ligand–protein interactions},
author = {Zhang, Chengxin and Zhang, Xi and Freddolino, Lydia and Zhang, Yang},
journal = {Nucleic Acids Research},
volume = {52},
number = {D1},
pages = {D404--D412},
year = {2023},
month = jul,
publisher = {Oxford University Press (OUP)},
doi = {10.1093/nar/gkad630}
}Details
Annotation Types
Domain
Use Cases
Tags
Found an issue or want to improve this design?
Open an Issue(opens in a new tab)Related Designs
ClinVar Variant Classification with ACMG/AMP Terms
Classify variants in igv.js with five ACMG/AMP terms and tick evidence criteria. The ten cases are illustrative, beside a live ClinVar track, and not a diagnosis.
BRACS Breast Lesion Regions on a Gigapixel Whole-Slide Image
Outline and label breast lesion regions with seven BRACS categories in the deep-zoom viewer, then train a tiled Mask R-CNN. The sample slide is illustrative.
Galaxy Zoo 2 Galaxy Morphology Decision Tree
Classify ten real Galaxy Zoo 2 galaxy images with an 11-question decision tree chained with display_logic, with a trainer pre-labeling the first question.