How to Download a Protein Structure from the RCSB PDB and Read the File

To download a protein structure, search rcsb.org by protein name, gene, or UniProt ID, open the entry, and check its experimental method, resolution, and R-free before you trust it. Click “Download Files” and pick PDB or mmCIF. Then read the file: ATOM lines are the protein, HETATM lines are ligands, ions, and water.

Every docking tutorial and molecular dynamics guide opens with the same silent assumption: that you already have a structure file on disk. That first step is where most beginners quietly go wrong. They grab the first search hit, keep the crystallographic waters by accident, miss that half a loop is not even in the file, and only find out when the docking box makes no sense or the MD run explodes on the first nanosecond. This guide, from the StemSkills Lab team (10+ years in sequence and structural bioinformatics, drug discovery, and molecular modeling), walks you through finding the right entry, judging its quality, downloading it, and reading it line by line.

What is the RCSB PDB, and why start here?

The Protein Data Bank (PDB) is the public archive of experimentally determined 3D structures of proteins, nucleic acids, and their complexes. The RCSB PDB (rcsb.org) is the main US access point to that shared archive. The original description still holds: the PDB is “the single worldwide repository of information about the 3D structures” of large biological molecules (Berman et al., Nucleic Acids Research, 2000).

As of 2024 the archive holds more than 200,000 experimentally determined structures (see the live RCSB holdings page), the large majority solved by X-ray crystallography, with growing numbers from cryo-electron microscopy and NMR. Every entry has a short identifier: a four-character alphanumeric code such as 1AKE. That code is how you cite, download, and reproduce work, so note it down the moment you pick an entry.

How do you find the right protein structure on the RCSB PDB?

Open rcsb.org and use the search bar at the top. Three searches cover almost every case:

• By protein or gene name: type something like “adenylate kinase” or a gene symbol. Expect many hits, because one protein can have dozens of deposited structures.
• By UniProt accession: if you already have a UniProt ID (for example P69441), paste it in. This ties the structure directly to a known sequence and is the most precise route.
• By PDB ID: if a paper already cites an ID, type it in directly.

One protein, many structures is the norm, not the exception. Different entries can be different organisms, mutants, ligand-bound versus apo (empty) forms, or different fragments. Use the left-hand facets to filter by organism, experimental method, resolution, and whether a ligand is bound. If your goal is docking a drug-like molecule, an entry that already has a co-crystallized ligand in the pocket is a strong starting point, because it shows you exactly where the binding site is.

How do you judge whether a structure is good enough?

Do not trust an entry just because it exists. Three numbers on the entry page tell you most of what you need before you commit.

Resolution is measured in Ångströms, and lower is sharper. Around 2.0 Å or better gives you confident side-chain positions for docking; above roughly 3.0 Å, side chains and waters get fuzzy. R-work and R-free measure how well the model fits the experimental data; R-free is the cross-validated version and the more honest one. A large gap between R-work and R-free is a warning that the model may be over-fitted. For X-ray entries, prefer the highest resolution structure of your target that still has the ligand or state you care about.

How do you download the file, and should you pick .pdb or .cif?

On any entry page, click Download Files in the top-right menu and choose the format. You can also pull files straight from the command line, which is how you will do it inside a real pipeline:

• wget https://files.rcsb.org/download/1AKE.pdb
• wget https://files.rcsb.org/download/1AKE.cif

The legacy PDB (.pdb) format uses fixed-width columns and is compact and easy to skim. The modern PDBx/mmCIF (.cif) format is the current master format of the archive and stores the same coordinates plus far richer metadata as labelled tables. The choice matters most for large structures.

Rule of thumb: use .pdb for a normal single protein because most beginner tools read it directly, and switch to .cif when the structure is a huge assembly that will not fit the old format. The wwPDB maintains the PDBx/mmCIF dictionary if you want the authoritative field definitions.

How do you read a PDB file line by line?

Open the .pdb in any text editor. Each line starts with a record type that tells you what it is.

• ATOM lines are the atoms of the standard biological polymer, your protein or nucleic acid. This is what you almost always want to keep.
• HETATM lines are everything else: bound ligands, ions, cofactors, and water. Water appears as the residue name HOH.
• Chain identifier (a single letter such as A or B) groups atoms into separate molecules. A dimer will have chains A and B; decide early which chain you actually need.
• altLoc, the alternate location indicator in column 17, flags atoms modelled in two positions (labelled A and B). Docking and MD tools usually want a single conformation, so you keep one set.
• B-factor (the temperature factor column) reports how much each atom “wobbles” in the model. High B-factors mark uncertain, flexible regions.

Two header lines catch most beginners off guard. REMARK 465 lists residues that are in the sequence but missing from the coordinates, usually disordered loops the experiment could not resolve. REMARK 470 lists residues with missing atoms. A structure with a gap in REMARK 465 has a physical hole in the chain, which matters if that loop sits near your binding site. Always scan these before modelling. The full column layout is defined in the official wwPDB file format documentation.

A few commands make this concrete on the file you just downloaded:

• Keep only the protein: grep '^ATOM' 1AKE.pdb > protein_only.pdb
• List the ligands and waters: grep '^HETATM' 1AKE.pdb
• Count NMR models: grep -c '^MODEL' structure.pdb

What should you keep or strip before docking or MD?

Reading the file is only useful if you then clean it deliberately. A safe default checklist:

• Pick one chain if the others are just copies of the same protein, unless the interface between them is your target.
• Remove crystallographic waters for most docking runs (grep -v 'HOH'), but keep any water that a paper shows is essential to binding.
• Decide on the ligand: strip a co-crystallized ligand to redock into an empty pocket, or keep it as a reference for where the site is.
• Resolve altLoc to a single conformation.
• Add missing atoms and hydrogens in your preparation tool, because most PDB entries do not include hydrogens.

Getting a clean, correct starting structure is the difference between a docking result you can defend and one you cannot. From here the workflow continues into preparation, docking, and simulation, which we cover step by step in the computational biology skills roadmap and the GROMACS molecular dynamics guides.

Troubleshooting: the traps that stump beginners

My NMR entry has 20 structures stacked on top of each other. NMR entries deposit an ensemble of models, each a valid solution. Most tools want one. Keep the first model (the block between the first MODEL 1 and its ENDMDL) or the representative model the entry page names, and delete the rest.

The molecule in my viewer looks wrong or incomplete. You may be looking at the asymmetric unit rather than the biological assembly. The asymmetric unit is what the crystallographer refined; the biological assembly is the functional form (for example, the full tetramer). On the entry page, download the biological assembly file if your target only makes sense as a complete oligomer.

My tool refuses to open the .pdb. Very large structures cannot fit the legacy PDB format because of the 99,999-atom and single-character chain limits. For ribosomes, viral capsids, and similar giants, download the .cif instead and use a tool that reads mmCIF.

Frequently asked questions

Is the RCSB PDB free to use? Yes. The archive is open and free, and structures can be downloaded and reused without a login.

What is the difference between the PDB and UniProt? UniProt is the reference for protein sequence and function; the PDB holds experimentally determined 3D structures. Many PDB entries are cross-linked to their UniProt record, so you can move between sequence and structure.

Do I need the .pdb or the .cif file? For a normal single protein, .pdb is fine and widely supported. Use .cif for very large complexes or when you want the complete metadata, since mmCIF is the current master format.

Why are there missing residues in my structure? Flexible or disordered regions often cannot be resolved experimentally, so they are recorded in REMARK 465 but left out of the coordinates. You can model them back in later if they matter to your question.

Can I use an AlphaFold model instead? Yes, for targets with no experimental structure, but treat predicted models with extra care: check the per-residue confidence scores and remember there is no experimental density behind them.