database searching

Question 1

BLAST

Answer 1

• very fast local search program
  
  • 50x faster than SW
• local is important
  
  • may only have one section/domain that is homologous
• finds ‘words’ in the query that have matches in the database
  
  • extend to form high scoring pairs that can’t be extended
  • evaluate statistical significance of hsp
• heuristic
  
  • tradeoff between accuracy and speed

Question 2

search sensitivity

Answer 2

• psiblast = iterative version of blast
  
  • increased sensitivity
  • accumulates similar sequences
• protein searches mroe sensitive
  
  • no similarity in nucleotide searches (same or not)

Question 3

blast output

Answer 3

• expect value and number of identities most important
• gaps also
• short alignments can have high identity but more likely to be due to chance
• longer sequence = higher bit score

Question 4

bit score

Answer 4

• re-scaled version of raw alignment score
• independent of search space size
• query sequence lenght n, sum of all database sequence lengths m, then search space is proportional to N = nm

Question 5

e value

Answer 5

• expected number of matches with same bit score or better that would be produced by random chance
• takes into account size of search space
  
  • bigger database increases chance of finding a match
• indication of real evolutionary relationship between query and match
• false positive rate
• greater than 1 indicates no statistical support for the relationship

Question 6

P value

Answer 6

• probability of obtaining a match by chance
  
  • different to E value (E can be greater than 1)
  • often the same

Question 7

multiple sequence alignments

Answer 7

• proteins form families with related sequences, structures and functions
• can learn more by aligning multiple sequences from related family members instead of pairwise alignments
• patterns of conservation reflect structural and functional evolutionary constraints
  
  • e.g. loops - less important for function, well conserved
• key functional residues often show strong conservation
  
  • e.g. ser proteases - conserved triad of ser, asp, his

Question 8

MSA algorithms

Answer 8

• more sequences so slower than pairwise alignments
• use heuristic methods
  
  • e.g. clustal

Question 9

clustal

Answer 9

• build guide tree
  
  • perform all pairwise alignments
  • group similar sequences
    
    • higher score → neighbouring branches
  • guides other pairwise alignments (more information)
• align sequences progressively
  
  • most closely related pairs aligned first
  • align next most closely related sequences to existing alignments

Question 10

MSA programs

Answer 10

• clustal omega
• T-coffee
  
  • slow, best for smaller alignments
  • align multiple MSAs
• MUSCLE
  
  • fast
  • estimates sequence similarity using short sequence words of n-residues
  • align small words and build up

Question 11

sequence profiles

Answer 11

• weight profile or PSSM
• captures information in MSA
• table of residue frequencies at each position of the alignment
  
  • can include gaps
• profile size nx21
  
  • number of sequences irrelevant
• only considers information in that set of sequences
  
  • similar rsidues not included are irrelevant
• can add to BLOSUM
• use to search PSI-BLAST for mor eremote homologues

Question 12

problems with frequency scores

Answer 12

• new sequence can differ from profile derived from existing set
• doesn’t include evolutionary knowledge of conservative subsitutions

Question 13

PSSM

Answer 13

• position specific scoring matrix
• substitution matrix (like BLOSUM) combined with observed frequencies

Question 14

PSI-BLAST

Answer 14

• position specific iterated blast
• form PSSM with MSA of blast search
  
  • closely related sequence sonly due to noise
  • search again with this profile
  • add more significant hits to refine profile
  • iterate until no more significant hits found (or for x iterations)
• bridges sequence space to find more distantly related homologues
• start with more conservative e value to ensure you get the right sequences

Question 15

PSI-BLAST

low complexity regions

Answer 15

• repeats of a few residues, often a coil
• limited information - unrelated protein sequences brought in
• need to be masked
  
  • SEG program in blast
  • run with and without - no clear default

Question 16

PSI-BLAST

drift

Answer 16

• sequences early on cna disappea rlater
• rogue matches can pollute the PSSM
• watch output and stop earlier
• or stricter E vlaue

Question 17

PSI-BLAST

asymmetry

Answer 17

• searching with sequence A may find B
• searching with B won’t necessarily find A

Question 18

HMMs

Answer 18

• statistically rigorous representation of protein families (via domains)
• markov chain
  
  • series of states connected by a transition probability
  • probability off state i+1 dpeends only on state i
  • probability of a sequence of states = products of probabilities of path that generated them
    
    • transition and output probabilities
    • delete, match and insert states (transition)