What is protein structure prediction?
Learning objectives
After reading this article you will be able to:
- Explain why protein structure matters for drug design
- Describe what AlphaFold and OpenFold predict
- Read the confidence scores that come with a prediction
A protein is a chain of amino acids that folds into a specific 3D shape, and that shape determines what it does and which molecules it binds. Protein structure prediction is the task of computing that 3D shape from the amino-acid sequence alone.
Why it matters for drug discovery
Structure-based methods such as docking need a 3D structure of the target. Experimental methods — X-ray crystallography, cryo-electron microscopy and NMR — are slow and do not work for every protein. Predicted structures fill the gap, covering proteins that have never been solved.
The AI breakthrough
For decades, prediction accuracy lagged far behind experiment. In 2020, DeepMind’s AlphaFold2 reached accuracy competitive with experimental structures for many proteins in the CASP14 blind assessment. Its successor, AlphaFold3 (2024), extends prediction to complexes of proteins with DNA, RNA, ions and small-molecule ligands.
OpenFold is an open-source, trainable reproduction of AlphaFold developed by an academic and industry consortium. OpenFold3 continues that work for biomolecular complexes, so researchers and companies can run, inspect and fine-tune the model themselves.
Reading the confidence scores
| Score | Meaning |
|---|---|
| pLDDT (0–100) | Per-residue confidence in local structure. Above ~90 is high confidence; below ~50 often indicates a disordered or flexible region. |
| PAE | Predicted aligned error between pairs of residues, in ångströms. Low values between two domains mean their relative position is reliable. |
| pTM / ipTM | Overall confidence in the fold, and in the interface between chains of a complex. |
Limits to keep in mind
- A prediction usually shows one static conformation; real proteins move, and a binding pocket may look different with a ligand bound.
- Side-chain positions inside pockets can be less accurate than the backbone, which affects docking.
- Predicted ligand poses and complexes should be checked against experimental data where possible.
Pyxis Discovery offers OpenFold3 structure prediction in the browser, with results viewable in 3D and usable for docking.
Last updated October 31, 2026
Put it into practice
Screen, dock and order compounds in one place
Pyxis Discovery combines macrocycle and small-molecule libraries with similarity search, docking and structure prediction in the browser.