Protein Data Bank

The Protein Data Bank (PDB) is the single global repository for three dimensional structural data of biological macromolecules, primarily proteins and nucleic acids. If you are a structural biologist, bioinformatician, drug discovery researcher, or graduate student learning molecular biology, this guide will help you navigate the PDB with a source-bounded and practical framework. Every structure in the PDB is experimentally determined, mainly by X ray crystallography, nuclear magnetic resonance (NMR) spectroscopy, or cryo electron microscopy, and it serves as the foundation for understanding molecular function, evolution, and disease mechanisms NCBI Bookshelf. The PDB is maintained by the worldwide Protein Data Bank (wwPDB) consortium, and its data are freely accessible. This guide covers core concepts, decision points, a step by step workflow, common mistakes, and limits of interpretation to ensure you use PDB data correctly and reproducibly.

At a Glance

Core Concepts and Decision Points

The PDB stores atomic coordinates, experimental metadata, and derived annotations for each structure. When you begin working with the PDB you must decide what kind of data you need: a single high resolution structure for docking studies, a set of structures from a protein family for evolution analysis, or a large collection for machine learning training EMBL EBI Training. Three core concepts guide these decisions.

Concept 1: The PDB ID and accession. Every entry has a unique four character alphanumeric ID (e.g., 4HHB for hemoglobin). This ID links all associated data, including sequence, ligand information, and publication. Always record the PDB ID in your methods section.

Concept 2: Resolution and quality metrics. Resolution (in angstroms) indicates how detailed the electron density map is. For X ray structures, a resolution below 2.0 Å is considered high, between 2.0 and 3.0 Å is medium, above 3.0 Å may be lower quality. For cryo EM, a resolution around 3.0 Å is now common but can range from 2.0 to 6.0 Å depending on flexibility. Always check the R free value (cross validation residual) and Ramachandran plot statistics (percentage of residues in favored regions) to gauge model reliability Galaxy Training Network.

Concept 3: Biological assembly vs. asymmetric unit. The asymmetric unit is the smallest unit that can be transformed by crystallographic symmetry to generate the full crystal. However, the biological function often involves a multimeric assembly (e.g., a homodimer). The PDB provides a "biological assembly" file when the biologically relevant oligomer differs from the asymmetric unit. Downloading the wrong one can lead to incorrect conclusions about binding interfaces.

Decision point: Which structure to use? If multiple structures exist for the same protein (e.g., with different ligands or mutations), select based on: (1) resolution (best quality), (2) completeness (no missing loops), (3) relevant bound ligand, and (4) experimental method that suits your question (e.g., cryo EM captures large complexes, NMR captures dynamics). Use the PDB validation reports provided by wwPDB to compare structures objectively.

Practical Workflow for Using the PDB

Follow this step by step sequence to extract and analyze PDB data reproducibly.

Step 1: Identify your target. Start with a protein name, gene symbol, or sequence. Use the search at RCSB.org or PDBe.org. Filter by organism, resolution, or experimental method. For example, searching "MMP9 human" yields dozens of structures, filter by resolution better than 2.5 Å and include the physiological ligand theaflavin if studying inhibition Toxicol Rep.

Step 2: Retrieve the entry. Download the mmCIF file (recommended over PDB format because it stores complete metadata, including all atom coordinates and experimental details). Also download the biological assembly file if needed. Many tools accept both formats, but mmCIF handles large structures (e.g., ribosomes) without truncation.

Step 3: Validate quality before analysis. Open the validation report PDF or XML for your PDB ID. Check the following flags: high number of Ramachandran outliers (above 2% is concerning), large clash score (above 20 per 1000 atoms indicates steric clashes), and poor side chain rotamer fit. For X ray structures, also check the R factor (expected below 0.25 for good data) and R free (roughly 0.06 0.10 above R factor). If the report shows problematic regions, consider using a different structure or modeling the missing parts with tools like Modeller Bioconductor.

Step 4: Extract relevant data. If you need coordinate subsets (e.g., active site residues only), use coordinate parsing libraries such as BioPython or ProDy. For large scale analyses, you can use the PDB archive in a compressed format (PDBx/mmCIF) and filter via command line tools. For sequence analysis, fetch the SEQRES sequence records, these correspond to the complete polymer chains, not just the modeled atoms.

Step 5: Visualize and analyze. Load the mmCIF file into molecular visualization software (PyMOL, ChimeraX, or VMD). Inspect the ligand binding site, hydrogen bonds, and interaction distances. For dynamic interpretations, remember that static PDB structures represent one snapshot. If your study involves conformational changes (as in millisecond timescale NMR dynamics), consider aligning multiple related structures to assess flexibility bioRxiv.

Step 6: Record provenance. Document the PDB ID, download date, file type, and any filtering criteria. This ensures reproducibility, especially when you combine PDB data with other resources like the NCBI Sequence Read Archive for structural genomics studies NCBI Sequence Read Archive.

Quality Checks and Common Mistakes

Common mistake 1: Using the asymmetric unit instead of the biological assembly. This can produce artefactual interfaces that do not exist in vivo. Always verify the oligomeric state mentioned in the PDB entry header or literature. The PDB provides a "Biological Assembly" file, use that for docking and interface analysis.

Common mistake 2: Ignoring ligand and cofactor information. Many entries include ligands, ions, or post translational modifications that are critical for function. For example, a study of Alzheimer disease plasma biomarkers uses structural information from PDB to interpret protein misfolding in prion diseases Neurology. If you analyze an enzyme without its cofactor, you may misassign catalytic residues.

Common mistake 3: Overinterpreting resolution. A structure at 3.5 Å resolution is not suitable for identifying water molecules or side chain rotamers, yet many papers still discuss specific hydrogen bonds without qualification. Use resolution to set expectations: at 3.5 Å, backbone traces are reliable but side chain positions are uncertain.

Common mistake 4: Not checking the publication. The PDB entry often links to a primary research paper that details experimental conditions, including pH, temperature, and ligand concentrations. Without this context, you cannot judge whether the structure is physiologically relevant. For example, a structure solved at pH 5 may not reflect the cellular environment at pH 7.4.

Quality check procedure: For any critical structure, download both the mmCIF file and the validation report. Check the clash score, Ramachandran outliers, and the deposited experimental data (e.g., electron density maps for X ray or 3D FSC curves for cryo EM). If the report indicates missing residues, note them in your analysis, never model them without justification.

Limits of Interpretation

Every PDB structure is a model derived from experimental data. It has intrinsic limitations.

Conformational states. A single crystal structure captures one stable conformation, often the lowest energy state in that crystal lattice. Proteins in solution sample multiple states, millisecond dynamics studied by NMR are not visible in most X ray structures bioRxiv. Thus, you cannot infer flexibility or allosteric pathways from a single PDB entry.

Missing regions. Many structures lack flexible loops, disordered N or C termini, or entire domains. These gaps are often labeled as "missing residues" in the SEQRES. Do not assume these regions are unimportant, they may be involved in regulation or binding. Use methods like AlphaFold or molecular dynamics to predict them, but note that those predictions are not experimentally validated.

Ligand bias. Crystal structures sometimes contain ligands or buffer components that do not occur in vivo, such as glycerol or heavy atoms from phasing. Check the PDB chemical component dictionary to distinguish between physiological ligands and experimental artifacts.

Evolutionary context. A structure from one organism may not represent the orthologue in another species. Sequence identity below 30% can lead to significant structural divergence. Align the sequences first, then evaluate whether the structure is a suitable template for homology modeling. For highly curated model organisms like the medicinal orchid Cremastra appendiculata, researchers have used PDB structures to annotate a de novo genome assembly Sci Data. Always cross reference with UniProt for domain annotations.

The resolution cliff. Cryo EM structures at near atomic resolution (2.0 3.0 Å) have become more common, but they often combine data from thousands of particles into one consensus map. Individual conformations are averaged out. For large assemblies, the overall envelope may be accurate, but side chain details for flexible subunits should be interpreted with caution. Similarly, NMR structures are ensembles of models, the average structure alone does not reflect the dynamic range J Chem Theory Comput.

Frequently Asked Questions

Q1: Can I use the PDB to confirm a protein protein interaction predicted by machine learning? A1: Yes, but only if a high resolution structure or a cryo EM map of the complex is available. Many interactions are predicted for complexes that have not been solved. Check the PDB for any complex containing both partners. If no structure exists, you can use models, but they are not experimental evidence. Cite the PDB ID if available, and clearly state if you used a predicted model.

Q2: What should I do if the PDB structure I need has poor validation metrics? A2: Look for an alternative structure of the same protein, possibly from a different species or solved under different conditions. If none exists, you may use the structure but report its limitations explicitly. In your analysis, discard residues that fall in high outlier regions. Some journals require deposition of re refined structures for reuse. Discuss with a structural biologist before basing conclusions on a problematic model.

Q3: How do I cite the PDB in my paper? A3: Cite the original research article that describes each structure (found in the PDB entry as the "primary citation") and also cite the PDB itself: Berman, H.M. et al. (2000) The Protein Data Bank, Nucleic Acids Research, 28(1), 235 242. Include the PDB ID in the methods section. Many journals also require the wwPDB validation report summary.

Q4: Is it acceptable to use cryo EM structures from the PDB for molecular docking? A4: It depends on resolution. For flexible or medium resolution cryo EM structures (above 4 Å), side chain placement may be unreliable. Docking programs require accurate atom positions. You can rigidly dock into the density map instead using tools like HADDOCK or Rosetta, which can incorporate cryo EM restraints. Always state the resolution limits in your docking protocol. For high quality structures (3 Å or better), docking is often feasible with appropriate preprocessing.

References and Further Reading

1. NCBI Bookshelf: Structure Databases , Free textbook chapter on molecular structure resources.
2. EMBL EBI Training: Protein Data Bank in Europe , Official tutorials for searching and analyzing PDB data.
3. Galaxy Training Network: PDB data analysis , Workflows for handling PDB files in bioinformatics pipelines.
4. Bioconductor: PDB and structural annotation packages , R packages for reading and processing PDB structures.
5. wwPDB Validation Reports , Official documentation on structure quality metrics.
6. Berman et al. (2000) The Protein Data Bank, Nucleic Acids Research , Original description of the PDB.
7. PDBj Training Resources , Tutorials from the Japanese node.
8. RCSB PDB: Learn , Educational material and video guides from the US node.

Related Articles

• Protein Synthesis
• Incomplete Dominance Gene
• Cell Membrane Function Biology
• Dna Structure
• Protein Structure