Biomarker = measured characteristics which may be used as an indicator of some biological
state or condition.
Homologous proteins often have similar structure and molecular function.
PSI-BLAST:
input = query sequence (the sequence you want to compare), database
output = for each potential hit: e-value, bitscore (alignment score), alignment, PSSM
parameters = reporting e-value threshold (if there was a hit or not), substitution matrix, gap
penalties, word size, inclusion e-value threshold for PSSM (if the hit should be concluded in
the building of the PSSM), max number of iterations (usually 6)
E-value = expected number of non-homologous sequences with score greater than or equal
to a score x in a database of n sequences. Smaller E = better. ex: E-value is 5, means 5 hits
have a score S (can be 5 different S's), which are all bigger than x, where x is the alignment
score of the sequence we actually found.
Bitscore = required size of database in which the current match could be found by chance.
It's a log2 scale and normalised raw score. Each increase by 1 doubles the required
database size (2^bitscore). It does not depend on the database. Gives the same value for
hits in databases of different sizes and hence can be used for searching in a constantly
increasing database. Higher score = better.
E = (m*n)/(2^bitscore)
Structure is better conserved than sequence.
PSI-BLAST basic idea: use results from BLAST query to construct a profile matrix, search
database, make conservation profile, search database with this profile instead of query
sequence, iterate this procedure.
synonymous mutation = DNA mutation that does NOT lead to an amino acid substitution.
nonsynonymous mutation = DNA mutation that DOES lead to an amino acid substitution: -
missense mutation (one AA replaced by another AA), - nonsense mutation (AA replaced by
stopcodon, what happens to the protein?)
Global pairwise alignment:
Find optimal score that matches all letters in both sequences. To find homologous
sequences, compare them and see which parts are more conserved.
input = 2 sequences, substitution matrix, gap penalties
output = optimal alignment score, sequence alignment (corresponding to optimal alignment
score)
, example of scoring algorithm: j is horizontal, i is vertical. So j-1 (horizontal) means
introducing a gap in the VERTICAL sequence, and i-1 (vertical) means introducing a gap in
the HORIZONTAL sequence. Going diagonal means +1 for a match, -1 for a mismatch.
(j-1 basically indicates that you move 1 place in the j direction! So horizontal in this case,
which introduces a gap vertically).
Lowest rightmost cell in the matrix is optical alignment score.
When tracing back and finding multiple optimal routes, choose the high road!
BLOSUM matrix:
Scores amino acids, so used for protein alignment, NO DNA. Which mutations are more
likely to occur?
BLOSUM N = made from conserved blocks in alignments with clustered with at most N%
sequence similarity.
+ values = preferred substitution
- values = avoided substitution
0 values = randomly expected
BLAST overview:
match query sequence with sequences in the database. Splice query into 'words' (substrings
of sequence)
Step 1. determining query words
Step 2. find 'near-exact' matches with scanning the database for occurrence of the words.
Matching of the words is scored by a given substitution substitution matrix. A word is
considered a match when it’s above a threshold
Step 3. extend the match. Stops looking when HSP begins to decrease
Step 4. local/pairwise alignment by extending from the words in both directions while
counting the alignment score using the same substitution matrix
Step 5: e-value
BLAST and PSI-BLAST are multiple alignment.
PSSM is a type of profile. For each sequence position, we can determine how likely it is that
a certain AA occurs.
A PSSM is ALWAYS as long as the query sequence (horizontal length)
A PSSM is ALWAYS 20 deep because there are only 20 amino acids (vertical depth)
(Unless the AA's are the columns… which is usually the case)
Profile drift = if you are too lenient with your E-value threshold, you'll include sequences that
are poor hits, which is going into your profile and is going to broaden to the point where you
start finding less and less relevant hits.
state or condition.
Homologous proteins often have similar structure and molecular function.
PSI-BLAST:
input = query sequence (the sequence you want to compare), database
output = for each potential hit: e-value, bitscore (alignment score), alignment, PSSM
parameters = reporting e-value threshold (if there was a hit or not), substitution matrix, gap
penalties, word size, inclusion e-value threshold for PSSM (if the hit should be concluded in
the building of the PSSM), max number of iterations (usually 6)
E-value = expected number of non-homologous sequences with score greater than or equal
to a score x in a database of n sequences. Smaller E = better. ex: E-value is 5, means 5 hits
have a score S (can be 5 different S's), which are all bigger than x, where x is the alignment
score of the sequence we actually found.
Bitscore = required size of database in which the current match could be found by chance.
It's a log2 scale and normalised raw score. Each increase by 1 doubles the required
database size (2^bitscore). It does not depend on the database. Gives the same value for
hits in databases of different sizes and hence can be used for searching in a constantly
increasing database. Higher score = better.
E = (m*n)/(2^bitscore)
Structure is better conserved than sequence.
PSI-BLAST basic idea: use results from BLAST query to construct a profile matrix, search
database, make conservation profile, search database with this profile instead of query
sequence, iterate this procedure.
synonymous mutation = DNA mutation that does NOT lead to an amino acid substitution.
nonsynonymous mutation = DNA mutation that DOES lead to an amino acid substitution: -
missense mutation (one AA replaced by another AA), - nonsense mutation (AA replaced by
stopcodon, what happens to the protein?)
Global pairwise alignment:
Find optimal score that matches all letters in both sequences. To find homologous
sequences, compare them and see which parts are more conserved.
input = 2 sequences, substitution matrix, gap penalties
output = optimal alignment score, sequence alignment (corresponding to optimal alignment
score)
, example of scoring algorithm: j is horizontal, i is vertical. So j-1 (horizontal) means
introducing a gap in the VERTICAL sequence, and i-1 (vertical) means introducing a gap in
the HORIZONTAL sequence. Going diagonal means +1 for a match, -1 for a mismatch.
(j-1 basically indicates that you move 1 place in the j direction! So horizontal in this case,
which introduces a gap vertically).
Lowest rightmost cell in the matrix is optical alignment score.
When tracing back and finding multiple optimal routes, choose the high road!
BLOSUM matrix:
Scores amino acids, so used for protein alignment, NO DNA. Which mutations are more
likely to occur?
BLOSUM N = made from conserved blocks in alignments with clustered with at most N%
sequence similarity.
+ values = preferred substitution
- values = avoided substitution
0 values = randomly expected
BLAST overview:
match query sequence with sequences in the database. Splice query into 'words' (substrings
of sequence)
Step 1. determining query words
Step 2. find 'near-exact' matches with scanning the database for occurrence of the words.
Matching of the words is scored by a given substitution substitution matrix. A word is
considered a match when it’s above a threshold
Step 3. extend the match. Stops looking when HSP begins to decrease
Step 4. local/pairwise alignment by extending from the words in both directions while
counting the alignment score using the same substitution matrix
Step 5: e-value
BLAST and PSI-BLAST are multiple alignment.
PSSM is a type of profile. For each sequence position, we can determine how likely it is that
a certain AA occurs.
A PSSM is ALWAYS as long as the query sequence (horizontal length)
A PSSM is ALWAYS 20 deep because there are only 20 amino acids (vertical depth)
(Unless the AA's are the columns… which is usually the case)
Profile drift = if you are too lenient with your E-value threshold, you'll include sequences that
are poor hits, which is going into your profile and is going to broaden to the point where you
start finding less and less relevant hits.
