Quiz Lecture 8

A protein has only 28% sequence identity to a known protein, but both adopt a very similar fold. What is the best conclusion?

• A protein has only 28% sequence identity to a known protein, but both adopt a very similar fold. What is the best conclusion?

• Structure prediction is impossible below 30% identity

• Proteins can still retain similar 3D structures despite low sequence identity

• The known structure must be wrong

Why is co-evolution useful in AlphaFold2?

• It directly measures binding affinity

• It helps identify residues that are likely close in 3D space

• It replaces the need for structural templates entirely

• It only works for membrane proteins

Someone says, “The MSA branch and pair branch in AlphaFold2 work independently.” Which reply is most accurate?

• Correct, they are completely separate

• Correct, they are merged only at the end

• Incorrect, the MSA representation can update the pair representation and vice versa

• Incorrect, but only in AlphaFold3

If AlphaFold2 predicts each domain of a protein with high pLDDT, what can you safely conclude?

• The relative orientation between domains is definitely correct

• Each domain is likely modeled reliably, but domain-domain positioning may still be uncertain

• The protein must be a rigid monomer

• PAE is unnecessary to inspect

Which situation best illustrates when PAE is more informative than pLDDT?

• Deciding whether a single helix is well formed

• Deciding whether a disordered tail is likely real

• Deciding whether two domains are confidently positioned relative to each other

• Deciding whether a sequence has homologs

A model has a long region with pLDDT below 50. What is the best interpretation?

• The region is definitely missing from the protein

• The region should be read as likely disorder rather than a reliable folded structure

• The full model is invalid

• • The sequence alignment failed, and some adjustments are needed

Why does AlphaFold2 use recycling?

• To reduce file size

• To replace the structure module and improve the Evoformer performance

• To feed outputs back through the network and improve prediction performance

• To generate ligands automatically

Which statement best captures the role of the structure module in AlphaFold2?

• It identifies evolutionary homologs

• It places residues in 3D space and predicts side-chain geometry

• It computes experimental electron density

• It scores protein-protein interactions in vivo

You want to predict a homomeric complex with many homologs from the same species. Which tool/setup is most promising?

• AF2-Multimer, because it tends to work well for homomers and large MSAs

• AF2 monomer only, because complexes are outside its scope

• RoseTTAFold only, because AF2-Multimer is unreliable for homomers

• Experimental methods only, because no predictor can handle this

A student concludes: “AF2-Multimer gave a confident interface, so these proteins definitely interact in vivo.” What is the best reply?

• Correct, a confident prediction proves the interaction

• Incorrect, because confident structural prediction does not prove biological interaction in vivo

• Incorrect, because AF2-Multimer never predicts interfaces well

• Correct, but only for heteromers