Structure Comparison and Classification

Question 1

ClustalW

Answer 1

• Positions specific gap opening and extension penalties
• Uses different amino acid substitution matrices - one for close relatives, one for distant relatives
• When the structure is known we want to increase the gap penalty within helices and strands and decrease it between them, forcing gaps to occur more frequently in loops
• If no structure is known, we can use simple rules which depend on the residues occurring and the frequencies of gaps

Question 2

Profile-based sequence search methods

Answer 2

• We can identify patterns of conserved residues by comparing related sequences within a protein family
• Even the most distant members of the family will have these patterns of conserved residues
• We can make a profile which encapsulates these patterns and use it to detect more distantly related sequences
• Highly conserved positions usually correspond to residues important for the folding or packing in the buried core or functional residues within the active site

Question 3

PSI-BLAST

Answer 3

• First, it constructs a multiple alignment of all the related sequences identified by BLAST
• Then it estimates the residue frequencies at each position to construct a position-specific score matrix (PSSM), also known as a weight matrix or 1D profile
• Then it uses the 1D profile for scanning the database

Question 4

HMMs for protein family recognition

Answer 4

• The sequence is aligned using a probabilistic model of interconnecting matches, deleted or inserted states
• Contains statistical information on observed and expected positional variation

Question 5

Protein Language Models (PLMs)

Answer 5

• Adapted from text learning - learn which features best predict the next residue in a protein
• These features can be represented in a vector for each residue to train classifiers for specific tasks
  e.g. ProtBERT, ProtTrans