Chemistry Labs

Biochemistry and biomedical chemistry

AI structure prediction (AlphaFold)

AI systems can infer protein structures from sequence patterns, related sequences and learned physical-structural regularities.

1 Lower secondary2 Upper secondary3 Undergraduate4 Advanced5 Research

IntuitionIntuition: the central idea

AI systems can infer protein structures from sequence patterns, related sequences and learned physical-structural regularities.

Explore the schematic model; it illustrates a concept rather than replacing experimental evidence.

SchoolSchool level: key concepts and a first application

Definition: Core concept

AlphaFold2 uses evolutionary information and neural-network representations with iterative structure refinement. Predicted confidence measures such as pLDDT describe model confidence, not experimental certainty.

Learned models combine multiple sequence alignments, coevolution signals and geometric priors. AlphaFold-style systems predict structures directly from sequence and now extend to complexes and ligands.

Key terms and meaning
TermInterpretation
MSAMultiple sequence alignment that exposes correlated residues.
pLDDTPredicted local distance difference confidence score.
EvoformerAlphaFold2 module refining pair and sequence representations.

Example: Apply the idea

A predicted segment has pLDDT 92 while a flexible loop has pLDDT 48. Which region is locally more confidently modeled?

Solution

The first segment is more confidently modeled according to pLDDT. This metric does not show whether the biological state, ligand, oligomer or conformation is correct.

UndergraduateUniversity level: quantitative description

pLDDT∈[0,100]\mathrm{pLDDT}\in[0,100]

Prediction quality varies by residue, disorder, ligand state and oligomeric context. Comparing predicted and experimental structures requires alignment and appropriate uncertainty interpretation.

AdvancedAdvanced: assumptions and mechanistic detail

Current systems extend beyond single chains to complexes, ligands and nucleic acids, but confidence is not uniform and physics-based validation remains essential.

ResearchResearch frontier: open questions and current practice

References