CAP/CGS 5991 Homework 2

Due: October 21, 2003.

Part A: BLAST and PSI-BLAST.

Your friendly biologist has asked you to retrieve the protein sequence of the E. coli RecA protein from SwissProt.

Q1: Explain briefly how you would find it. If you type in RecA Ecoli as the search text, you may not find it. Although you may not use this information for the search, the primary accession number for this sequence is P03017. This information is merely for you to make sure that the protein you found is the correct one. Your friend also tells you that its GenBank accession number may be gi:72985. Next try to locate this protein with the accession number gi:72985 from the Entrez database browser and answer the following questions.

Q2: What is its correct gi number? How long is the protein sequence?

Q3: Now run it through BLASTP, like you did in the first assignment and answer the following questions.

  1. Write down the organsims and the gi numbers of the top three hits.
  2. What scoring matrix and gap penalties were used?
  3. What value of K and l were used for calculating the Expect scores for the gapped alignment (please note that there are two sets of these paramaters - one for ungapped and one for gapped alignments)? Where do these values come from?
  4. The score shown in the program output is in units of "normalized bits" = [(l x raw score) - ln K] / ln2. The raw score is shown in parentheses. What are the units of the raw score (those of the BLOSUM62 matrix)? Calculate the raw score in bits from the "normalized bits" for the top three hits.
  5. How many database sequences were searched?
  6. Is the alignment of the highest scoring sequence with RecA protein significant and why? What biological information (protein structure and function) does this match suggest about the bacterial RecA protein and the top three hits?
  7. What was the lowest reported score in this search, and is this score significant?
We are now going to try PSI-BLAST. Read about it by going to following tutorial (Click here). PSI-BLAST is a version of the BLAST algorithm that uses the results from an initial search for similar protein sequences to construct a type of scoring matrix that can then be used for additional rounds of searches, called iterations. The variability found in each column of the scoring matrix allows additional sequences that have different combinations of amino acids in the sequence positions to be found. The algorithm provides a rapid but less precise search than other methods because the scoring matrix produced is only approximate and includes most of the original query sequence. (Caution: The iterations can lead to more sequences being added that do not share a region in common with the original query sequence, but share a totally different region in some of the added sequences; e.g., these new sequences are not true family members but foreigners.) The process will stop when no more sequences are found. The user can control the number of sequences to be included at each iteration or else use the score cutoff recommended by the program. The method is often used to perform a rapid and preliminary search for members of a sequence family. The found sequences can then be multiply aligned by other better-defined methods. First go to the GenBank page for MITF_MOUSE (gi|13124350|sp|Q08874|). Download it in FASTA format and run it on PSI-BLAST (go to BLAST and click on the PSI-BLAST option). Use the following changes to the default values (or else you will get way too many hits). At the bottom of the page, under "Format" option, look for the options for "Number of: Descriptions" (Pick 1000 here) and "Alignments" (Pick 0 here). For the options for "Limit results by entrez query or select from:", pick "Mus Musculus".

Q4: You should have gotten 88 hits. What default threshold was used for the E-value? How many of these sequences had E-value BETTER than the default threshold? Run PSI-BLAST for two more iterations. How many more hits did you get?


Part B: Multiple Alignment.

CLUSTALW is a widely used multiple sequence alignment tool. This assignment will expose you to the features and capabilities of this program.

  1. Go to Entrez
  2. You need to download the protein U1A from four different organisms (human, mouse, Xenopus laevis, and Drosophila melanogaster). For some of these organisms, you may find several proteins being found by Entrex. Pick the ones with the following accessions numbers: human - 2554638; mouse - 543325; Xenopus laevis - 65181; Drosophila melanogaster - 1173325. You can verify that you have the correct sequences by looking in the GenPept Report for the sequences you picked. For each of the four sequences, open the FASTA report and copy them all into one file.
  3. Go to CLUSTALW.
  4. Type in your e-mail address. Either upload your file or paste your sequences into the appropriate box. Now run CLUSTALW.
  5. Study the output that you get within a few moments.
  6. Try the "JalView" option.
  7. Go back to the CLUSTAL page, and try it again after changing some of the settings. Try a different substitution matrix. Also try different gap penalties.
  8. JalView has an option to mail yourself the postscript version of the alignment. Try this option.
  9. Go back to the CLUSTAL page, and change the "TREE TYPE" to "phylip". Look at the tree that is output.
  10. Pick 4 more (new) sequences that are related to the U1A proteins (run it through BLAST and pick 4 hits with good E-scores). Now align all 8 of the sequences using the "TREE TYPE" of "phylip".
How long are the 4 sequences? What are the pairwise alignment scores?

Q1: What do the "*", ":", and the "." in the alignment indicate? Consult the substitution matrix values, if necessary. Were there any differences in the alignment when you tried two different substitution matrices (PAM and BLOSUM)?

Q2: What sequence formats are supported by CLUSTALW?

Q3: How can the tree information be interpreted? Can you draw the tree that you obtained when you used the "TREE TYPE" of "phylip" with the 8 sequences that you aligned? What do the numbers in the tree information mean? You can download your own version of CLUSTAL (called ClustalX) from ftp://ftp-igbmc.u-strasbg.fr/pub/ClustalX/ Download the appropriate version for your machine (for MS Windows, it is called "clustalx1.81.msw.zip"). Versions for other operating systems also exist.

Q4: Download ClustalX for your personal machine and write down the differences you see in the alignment (for the same input as before) from the Web-based ClustalW version you used above. You can visualize the tree, assuming you also downloaded the tree visualization components. Print out the resulting trees in one of the four formats.


Part C: Hidden Markov Models.

Here is a small alignment of 12 members of a DNA sequence family.
(column:  1234)
seq1      GATC
seq2      CTAG
seq3      GATC
seq4      CC-G
seq5      GATC
seq6      CC-G
seq7      GTAC
seq8      CG-G
seq9      GCGC
seq10     CTAG
seq11     GATC
seq12     CTAG
Suppose you were to build a profile HMM of this alignment. The profile has four match states; match state 1 is assigned to the symbols in column 1, etc.

Q1: Draw a profile HMM in terms of states (circles) and state transitions (arrows). You need to use the "Learning Algorithm" we discussed in class for HMMs. Note that unless you remove states that have no probability of being reached from the "Begin" state, you will be unable to work out this problem by hand.

Q2: Calculate the emission probability parameters for A,C,G,T in match state 1 (column 1). Do a maximum likelihood estimate, i.e., ratio of the frequency of that character being emitted to the sum of frequencies of all the characters.

Q3: Using the above answer, calculate the "log odds scores" (equal to the log of the ratio of its emission probability to its background frequency) for A,C,G,T in match state 1. Assume that the expected background frequencies of A,C,G,T are each 0.25. Use log base two so your scores are in units of bits.

Q4: Column 3 has gap symbols which would be assigned to delete state 3. Calculate the scores (log_2 probabilities) for the match_2 -> match_3 state transition and the match_2 -> delete_3 state transition.

Q5: Calculate the HMM log odds score (in bits) for the sequence

 GAAG 
and the sequence
 GATC
Notice that columns 1-4 and 2-3 covary as if they are Watson-Crick base pairs. It would therefore seem that the sequence GAAG should not be a true member of the sequence family. (Hint: the score will be the sum of four emission log-odds probabilities and one state transition log probability, since all other state transitions have probability one in this case. Also, make the Viterbi assumption that the obvious alignment of the four symbols to the four match states is correct, so you do not need to sum over all possible paths.) Now recall the discussions we had in class about the disadvantages of HMMs for the next question.

Q6: Is the HMM a good model of the pairwise correlations? Comment on the limitations of the HMM model.

Q7: [Extra Credit] How can you modify the HMM model so that it recognizes the correlation between locations? It may help to first ignore the correlation between locations 2-3 and only assume that locations 1-4 have a correlation.