PSI-BLAST website algorithm parameters?

Question

http://blast.ncbi.nlm.nih.gov/Blast.cgiIn this website, when I want to apply the psi-blast algorithm on a sequence, under the section of algorithm parameters , what does PSI-BLAST threshold mean? The question mark says it means that:>Set the statistical significance threshold to include a sequence in the model used by PSI-BLAST to create the PSSM on the next iteration.Another parameter is there called pseudocount. Did not understand that too. Could not understand what this means, someone please explain in simple language.

Answer 1

Threshold - decided which sequences get into the model

PSI-BLAST threshold values (E values)

E value measures how likely a sequence is to occur by chance.

smaller number = strict, only a few reliable hits (0.0001) - could miss real homologs

default = 0.005

larger number = looser, many hits (0.1 etc.)

PSI-BLAST first runs like a normal BLAST search.

1. It finds matching sequences (hits).
2. Trusted hits (based on the threshold) are aligned into a multiple sequence alignment (MSA).
3. The MSA is used to build a PSSM (position-specific scoring matrix).
4. PSI-BLAST searches the database again using the PSSM instead of a general scoring matrix.
5. This makes the search more sensitive.
6. New trusted hits may be added.
7. The PSSM is updated.
8. Another iteration begins.

Pseudocount - prevents PSI-BLAST from assuming something is impossible just because it was not observed.

Pseudocount - is a small extra value added so PSI-BLAST doesn’t assume something is impossible just because it wasn’t observed.

Low pseudocount = very specific, sensitive to noise, position 10 = A for all 5 sequences

High pseudocount = smoother, more conservative profile, position 10 = A for all 2 sequences

This isn't a fixed number. It is calculated based on:

(1) number of aligned sequences at each position

(2) how conserved or variable that position is

(3) background amino-acid frequencies derived from known databases.

Example:

Suppose PSI-BLAST aligns 5 sequences at one position: position 10 = A for all 5 sequences

Observed counts (no pseudocount):

A = 5

All other amino acids = 0

Probabilities without pseudocount:

P(A) = 1.0

P(any other amino acid) = 0

This means PSI-BLAST would treat all other amino acids as impossible at this position.

Same example with a pseudocount applied:

Adjusted counts (conceptual):

A = 5 + small extra

Other amino acids = 0 + very small values

Example probabilities with pseudocount:

P(A) ≈ 0.90

P(G) ≈ 0.005

P(S) ≈ 0.005

P(other amino acids) ≈ small but nonzero

This means A is strongly preferred, but other amino acids are still possible.