prediction of protein function

Question 1

sequence similarity

Answer 1

• most common method of function prediction
  
  • high identity indicates similar function
• but proteins with similar sequences can have different function
  
  • one position changing can change funciton e.g. binding
• experimental confirmation

Question 2

MSAs

Answer 2

• show where to look for evolutionary constraints
• highly conserved residues important
• e.g. pml

Question 3

pml

Answer 3

• human TF with RING domains
  
  • zinc finger protein
• Zn coordinating residues for binding
  
  • C..H..C..C
    
    • found in MSA

Question 4

regular expressions

Answer 4

• characteristic sequence patterns in PROSITE
• more plasticity than a consensus sequence

Question 5

sequence logos

Answer 5

• more detailed representation than regular expression
• total column height (bits) = degree of conservation
• height of each symbol = relative frequency of that residue in that position

Question 6

profile HMM

Answer 6

• needed to find more remote homologs
• machine learning approach to funciton prediction
• profile = based on MSA
• hidden = states not directly observed
• markov = state depends only on state before

Question 7

Pfam HMMs

Answer 7

• MSA of full domain
• characterise columns on conservation
  
  • absolute/high conservation (capital)
  • some/no conservation (lower case)
  • insert (generates new column)
• conserved columns make up a consensus

Question 8

HMM states

Answer 8

• insert
  
  • generate a residue for an inserted column
• delete
  
  • remove a normally conserved column
• match
  
  • generate a residue for a conserved column
  • according to residue frequencies in MSA
• HMM machine for generating sequence belonging to that protein family
  
  • probabilistic model
  • each path through model has associated probability

Question 9

forward algorithm

Answer 9

• dynamic programming in pfam
• finds probability that model could generate a given sequence
• align HMM with sequence
  
  • get likelihood for the sequence-profile alignment
  • bit score and E value
  • identities and similar residues
  • posterior probability

Question 10

targetP

Answer 10

• profile HMMs to identify signal peptides characteristic of specific cellular compartments
• mitochondrion, chloroplast or secretory system

Question 11

TMHMM

Answer 11

• HMM to detect TM helices
• predict architecture of integral membrane proteins

Question 12

STRING

Answer 12

• uses multiple sources of information to predict functional interactions
  
  • genomic context
  • high throughput experiments
  • coexpression
  • previous knowledge
• some aspects of function not directly related to sequence

Question 13

sources used by STRING

Answer 13

• genomic context:
  
  • conserved proximity of 2 genes in the genome
  • e.g. trpA trpB
    
    • can sometimes fuse into one protein coding gene
  • indication of involvement in same pathway
• confirm with protein interaction experiments
• gene coexpression experiments
  
  • conserved expression pattern indicates proximity
• text mining
  
  • search abstracts for co-mentions