protein structure prediction · BEGINNER LESSON
AlphaFold
AlphaFold is a deep learning system that predicts a protein’s three-dimensional structure from its amino-acid sequence. It is used to estimate a likely folded shape, not the folding pathway or all possible states of the protein.
What it is and what it measures
AlphaFold is an artificial intelligence system for protein structure prediction. Its input is a protein’s amino-acid sequence, which is the line of building blocks in a protein, and its main output is a predicted three-dimensional structure. It can also provide uncertainty information about the relative positions and orientations of residues, the individual building blocks in a protein chain. [0,2]
[1] AlphaFold and protein folding: Not dead yet! The frontier is ... [3] Great expectations – the potential impacts of AlphaFold DB
How it works
The system was trained on many experimentally determined protein structures and large amounts of protein-sequence data. By learning relationships between sequence and known structures, it can make predictions for sequences it has not seen before. In practice, this makes AlphaFold useful as a fast starting model for structural biology work. [0,2]
[1] AlphaFold and protein folding: Not dead yet! The frontier is ... [3] Great expectations – the potential impacts of AlphaFold DB
What AlphaFold does not tell you
AlphaFold usually predicts one structure, while real proteins may shift among many conformations, meaning different shapes they can adopt. It does not provide the mechanism of folding, does not sample the full distribution of shapes, and is not trained to predict the effects of mutations. It also does not predict protein-protein, protein-DNA/RNA, or protein-ligand complexes, and it performs poorly for intrinsically disordered or flexible regions. [0,1,2]
[1] AlphaFold and protein folding: Not dead yet! The frontier is ... [2] Advantages and Limitations of AlphaFold in Structural Biology - PMC [3] Great expectations – the potential impacts of AlphaFold DB
Why the limitations matter
Because proteins can change shape with environment, binding partners, or functional state, a single predicted structure may miss important biology. Experimental methods such as cryo-EM, cryo-ET, and X-ray crystallography are still needed to validate predictions, resolve complexes, and study state switching or flexible regions. AlphaFold is best treated as a powerful auxiliary tool, not a complete replacement for experiments. [0,1,2]
[1] AlphaFold and protein folding: Not dead yet! The frontier is ... [2] Advantages and Limitations of AlphaFold in Structural Biology - PMC [3] Great expectations – the potential impacts of AlphaFold DB
SEE THE IDEA
AlphaFold in structural biology
The protein’s starting linear code
Learns sequence-to-structure patterns from known proteins
A single likely structure with uncertainty estimates
cryo-EM, cryo-ET, or X-ray crystallography can check or refine the model
Sequence is the input; AlphaFold predicts a structure; experiments are used to test, refine, or complement that prediction.
AlphaFold connects sequence data to a structural model, but experiments are still needed when proteins are flexible, complex, or context-dependent.
TRY IT YOURSELF
Test your understanding
Choose an answer, then check the explanation.
PUT IT TOGETHER
Match each concept to its meaning
Drag a label onto its description, or select a label and then select a description. This also works with a keyboard.
4 pairs to match
Sources and further reading
This general lesson was generated from public source material. It explains a concept; it does not interpret a private paper or provide medical advice.