Skip to main content
Loading Explore guides…
Bioinformatics · Acedra
Home
Explore
Biology
Bioinformatics
Sign in to save
Chat
If YouTube asks you to sign in, open the video on YouTube.
Watch on YouTube
Reload player
Transcript
161 segments
Copy
Hide video
0:02
Andy Baxevanis: All right, good morning, everyone, and welcome to week two or this year's current topic series. Let's go ahead and get the formalities out of the way right from the start. So my CME disclosure statement, "I don't have any relevant financial relationships with any commercial interests." Okay, so with that out of the way, my job today in the same way that Dr. Green laid the groundwork for upcoming lectures
0:24
in this series by providing you a broad overview of the field of genomics last week, I'm going to be providing you a solid foundation for the upcoming lectures devoted to bioinformatics. I mentioned last week that genomics and bioinformatics go hand in hand with one another; they're inexplicitly linked. And Eric really reinforced that sentiment when he showed you this figure from NHGRI's current vision document, showing the spectrum of genomic-based research going
0:48
from the understanding of the structure of genomes all the way through to our efforts focused on improving the effectiveness of healthcare. Now, our ability to work across this entire spectrum really rests, in large part, on being able to apply computational approaches across the board. To analyze the kinds of data that are generated in the course of whole genome sequencing efforts, genome-wide association studies, clinical sequencing efforts,
1:15
and pharmacogenomic studies, or any of the other many kinds of genomic science you'll be hearing about throughout this series. So given, again, that inextricable relationship between genomics and bioinformatics, what we're going to go through over the next two weeks is as gentle of an introduction as I can give you to some of the major tools used in bioinformatics. You're going to see the tools and the concepts underlying these tools again and again throughout this series. So
1:42
it's really important to have a basic understanding of the theory that underlies these techniques to start taking this technique -- these techniques away from the realm of the black box. So I hope you'll find this information useful, and that you'll pick up a few strategies along the way that will hopefully be of practical use in your own studies. So with that, let's just start right off by reinforcing the importance of alignment as
2:06
one of the most powerful approaches that we have in the study of sequence data, specifically with respect to determining similarity and deducing homology. So start off with a very basic question: why bother constructing sequence alignments; what do they do for us? And that question applies whether we're talking about a paralyzed [spelled phonetically] alignment, where we're taking two individual sequences, and aligning them to one another, or to a
2:29
multiple sequence alignment, where we're looking at a collection of presumably related sequences. So in either case what we're trying to do is get a measure of relatedness between either nucleotide or amino acid sequences, and once we have that measure of relatedness it allows us to draw certain biological inferences. So we might find structural relationships between these sequences, functional relationships, or evolutionary relationships.
2:53
Now, before we start to talk about the technique we need to define a couple of key terms; it's important to use the right terminology when we talk about these phylogenetic relationships. So the first term we're going to talk about is similarity. So what I'd like you to do is imagine you have two sequences, A and B, and that you align those two sequences to one another. And the easiest way to convey how related those sequences are to one another
3:17
is to just simply count the number of matches and mismatches along that alignment. So that gives you the quantitative measure; again similarity: it's based on an observable -- in this case how many positions are aligned with one another? We usually do express this as a percent identity, and by looking at this it allows us to quantify changes that we see as two sequences diverge from one another, whether we have substitutions, insertions,
3:42
or deletions. It'll also, from a functional standpoint, allows us to identify residues that are crucial for maintaining structure, and by extension, function. If we start seeing very high percentages of sequence similarity, high degrees of sequence similarity, what that might be telling us is that these two sequences in this paralyzed alignment have a common evolutionary history, or possible commonality in biological functions. So for
4:11
example, we might be able to identify regions that are important from a functional standpoint when thinking about things at the protein level, or regions that are important to expression when we're looking at sequences at the nucleotide level. The second term we need to define this morning is the one that is most often misused, and that one is homology. So what homology is, is the conclusion that we get by looking at
4:35
similarity; it implies an evolutionary relationship. When there is homology between sequences those sequences are termed homologs. So these are genes that have arisen from a common ancestor. And unlike similarity -- that is the quantifiable term -- this is the conclusion, again. So either genes are, or are not homologous; we don't measure this in degrees. And just to drive home the point, a quote from Walter Fischer who was one of the grandfathers of
5:02
bioinformatics: he says -- it's worth repeating here, that "homology, like pregnancy, is indivisible; you're either homologous/pregnant, or you are not." So again, this is the conclusion. Now, that term, homolog, actually breaks down into several subtypes. It might describe several different types of evolutionary relationships, and we're going to focus specifically on two of them here. So the first type of homology has to do with orthologs. So orthologs are
5:27
genes that have diverged as a result of a speciation event. So they're direct descendants of a sequence in a common ancestor, they more -- likely have similar domain and three-dimensional structure; because of that, they usually retain the same or very similar biological functions over evolutionary time, and we can use that information to predict gene function in novel genomes that we're starting to assign functions to newly identified genes. So put very simply
5:54
orthologs can be thought of as the same gene in different species. The second term, the second type of homology we're going to talk about, is paralogy. So when we have paralogs, these are just genes that arose by the duplication of a single gene in a particular lineage. So these might be less likely to perform similar function, but they can take on new functions over evolutionary time. So this is where Mother Nature gets
6:18
to be inventive, providing some insight into evolutionary innovations; taking bits and pieces of proteins, restructuring domains in order to accomplish new jobs within the cells. So again, put simply paralogs can be thought of as similar genes in the same species, adapting or co-opting a preexisting gene product for new biological functions. So just to make this a little bit clearer, because the definitions are a little hard
6:45
to follow, but so let's look at a picture. So in this case we have a gene, gene alpha, in a common ancestor and along the -- oh, evolutionary time there's been a couple of speciation events that have given rise to three species. Each one of them has an ortholog to gene alpha, Gene 1, Gene 2 and Gene 3. So each one of these three genes arose from this common ancestor alpha. Now, if we go back a little further in evolutionary time,
7:15
we roll back the clock a bit, and we consider the possibility of a gene duplication predating alpha. So we have a gene duplication here at the base of the tree giving rise to genes alpha and beta in this case. So alpha went along and gave rise to three orthologs; beta in this case underwent some similar speciation events -- not exactly the same -- but gave rise to three genes, 4, 5, and 6. So, how are these things all related to one another?
7:43
Any of the Genes 1, 2, and 3 are orthologous to one another; Genes 4, 5, and 6 are also orthologous to one another, because again, in each case they had a common ancestor. But now, if we take any pair of 1, 2, and 3 versus 4, 5, and 6, those are the paralogs, because they go way back to this gene duplication event; that's how they're related to one another. So that takes care of the definitions. Now, how do we actually identify the orthologs,
8:12
the same gene in different species? And the most common way, and most straightforward way we have of identifying candidate orthologs is by looking for reciprocal best hits. And when I say candidate orthologs, it's very important -- the word candidate really, really is important, because, keep in mind that a lot of the things we're going to talk about through this series deal with computational predictions, so I really need to underscore
8:36
the importance of being able to experimentally verify a lot of the things that you do using these computational techniques. Okay, so let's say we start with a gene that we're interested in from a particular organism,
8:51
and we -- as a first step here, we just do a sequence similarity search against all of the genes in a second organism. And what we might find is something that looks like this, a series of hits using some sort of sequence similarities search method that identifies a series of genes that are similar to the one that we started with. Now, I have the first one here marked in yellow, so we're going to just assume that based on some scoring
9:20
method, this one is the most related to the red one that we started with. Okay, so that's the first step. Now, where does the reciprocal part come in? Now, what you do is you take this yellow sequence and do the sequence similarity search in the opposite direction. You go -- now take this, and search it against all of the genes in Organism 1. And if you're lucky, what you'll find is that the best hit in both directions involve these two genes. So the
9:44
red one is most related to the yellow one in Organism 2; then, when you use the yellow one as the search query, it's most related to the red one in Organism 1. Okay, so that's the general idea. Now for the specifics. We have two broad types of alignment method at our disposal. So the first one are called global sequence alignments. So what I want you to do is imagine that you have two full-length sequences, either two
10:11
genes or two protein sequences, and you now align them across their entire length. So we're going to force an alignment over the entire length of these sequences. So this works really well if you have sequences that are of similar length, but the problem now is that, as these become more divergent, as they become more dissimilar to one another, the global alignment methods tend to miss important biological relationships; because
10:38
you're trying to force an alignment over the entire length of these sequences, it might mean that you're just not aligning important structural or functional domains. So from a biological point of view, a slightly different approach is better, and that's what are called the local sequence alignments. So here, we try to find the most similar regions in two sequences that are being aligned with one another, and the glorified term for this
11:01
is paired subsequences; it just means that we're just looking for two regions within an entire sequence that align with one another. So any regions, when we do these local sequence alignments, outside the area of the local alignment, we just go ahead and exclude those. By doing this, we're aligning two sequences; we might find more than one local alignment that could be generated for any two sequences that we compare. So this works really well
11:27
for sequences that share some similarity, but not high degrees of similarity, or for sequences of different length. So again, I'll just emphasis that is really the best way to compare sequences from a biological standpoint, especially if we're trying to deduce evolutionary relationships. So the approach of choice for biological discovery. Okay, so let's say we have now aligned two sequences, we talked about the concept of
11:52
similarity we're going to try to look for percent identity, some metric to allow us to quantify how good the alignment actually is. How do we actually deduce that number, how do we measure how good the alignment is between Sequence A and Sequence B? So to do that we use something called scoring matrices, and these scoring matrices quite simply is our empirical weighting schemes that we use to assign these numbers, these numerical values.
12:20
And what's in those scoring matrices is information concerning the physical, chemical, and biological characteristics of either nucleotides or amino acids. So they take into account things like side chain structure, their chemistry, the function of particular side chains based on what we know about how -- especially on the amino acid side of the house -- how these particular amino acids actually are involved in structure and function. So we know that
12:44
things like cysteine and proline residues help to keep parts of a protein together; tryptophan has a bulky side chain, so it can only be accommodated in certain parts of a protein; things like lysines and arginines, they're positively charged, they're very important, particularly in the context of DNA binding proteins, because you need those positively charged residues to bind to the negatively charged backbone of the DNA.
13:08
Okay, so in the construction of these scoring matrices, we try to take into account two major factors. First is conservation: which of the residues can substitute for any other residue, and not adversely affect the function of the protein? So we know that things like isoleucines and valines can substitute for one another in a protein sequence, because they're both small and they're hydrophilic. Things like serines and threonines, they're
13:33
both polar so we often see that substitution. So the basic idea here is we want to just keep track of charge, size, hydrophobicity, and other physical, chemical factors, again, to try to see, when we can't have an exact match between two sequences, what's the next best thing? The other factor is frequency. If we look at the entire constellation of proteins, all of the protein sequences available to us, how often do we see any one of those
13:59
amino acids amongst the entire constellation of proteins? Some are more rare; some are more frequent, and that becomes important. Okay, so why should you actually care about any of this? Every single time you do any sort of sequence comparison, whether it's two sequences to one another, A to B, or a Sequence A to an entire data base, like GenBank, these scoring matrices are being used in the background; they appear in every single analysis involving sequence comparison. They also,
14:29
as you'll come to hopefully appreciate, they implicitly represent evolutionary patterns. So the choice of matrix really can strongly influence what results come out of the methods that you apply to these alignments. So the default option may not always be the best choice, and so it's -- becomes important for us to talk about how these scoring matrices are actually constructed. What guidelines go into putting them together?
14:55
All right, so let's start on the easier side of the house. If we look at a scoring matrix that -- if we're doing a nucleotide-based comparison -- we only have four nucleotides A, T, G, and C -- and if we align two sequences with one another, all we're going to do is just apply a very simple matched or mismatched scheme. So anytime you have two -- a particular position in an alignment where the two nucleotides match which one -- one another, you're going
15:22
to assign a positive value; any time you have a mismatch between the two, you're going to subtract points. So in this case, the match gets you two points; the mismatch loses you three. The matrix, as you'll see here, you have the A, T, G, and C going across the top, the same residues going down the side, and just the way it works, that for every position in the matrix -- let's say you have a T aligned with a T -- you just look for the intersection,
15:44
and you'll see that that's -- in this case it's an exact match, so that's the plus two. You'll see the plus twos all across the diagonal representing the exact matches, and the minus threes in the other positions. This assumes that each nucleotide occurs 25 percent of the time. Okay, so this works really well at the nucleotide level, because nucleotides are not all that chemically dissimilar from one another, but as we've already alluded to, the new amino
16:12
acids are very different from one another, a much more complex picture. So we now go to something that looks like this. So this matrix is called a BLOSUM62 Matrix, and again, we have all of the amino acids going across the top, all of them going down the side. So again, I want you to imagine, now, in this case a protein sequence alignment, Sequence A with Sequence B. We're going to look at a particular position, and see how many points
16:38
we can assign to a match at that position. So we're going to image where we've got a tryptophan in Sequence 1 and a tryptophan in Sequence 2, so we're going to look for where the W intersects with the W, and that give us in this case 11 points. Now, if you look at the diagonal here, unlike the case we saw before, where no matter which nucleotides we were aligning and getting exact matches, where in the previous case we were assigning
17:02
the same value regardless, here on the diagonal you'll see that the values change. The 11 is the highest number in the diagonal, and that represents the frequency part of constructing the matrices. This is the most rare of the amino acids, so when you get an exact match you want to give it more points than you would if you had a more common amino acid aligning with each other. Okay, now to the point of substitution. All right, so let's say, instead of having an
17:32
exact match of tryptophan for tryptophan, this time we've aligned a tryptophan with a tyrosine. So instead of W-W, we now have a W against a Y. We take a look here, we see now in this case we would only give it two points. So it's still a positive score, which represents that it's a conservative substitution -- those two things can substitute for one another, and in most cases not adversely affect the structure or function of that particular
17:56
protein -- but the lesser value here represents that it's just not as good as an exact match. Let's just take as a final example a mismatch. So let's say you've now aligned the cysteine with a glutamine. Those two things are not known to be able to substitute for one another, so here we are to subtract points; in this case, minus three. So the way the general rules work, just in summary, you get the most points for an exact match with the rare amino
18:23
acid scoring more points than the common amino acids; you lose points for conservative substitutions, and you lose points for mismatches. Okay, now, so the matrix we were just looking at was something called a BLOSUM62 Matrix. It is part of a family of matrices called the BLOSUM Matrices. These we're developed by the Henikoffs at the Fred Hutchinson Cancer Research Center. And that's part of the good news: you don't have to derive these matrices yourself; you just need to know which ones
18:51
to pick. So as far as the BLOSUM Matrices go, what they were designed to do is look for differences in conserved un-gapped regions of our protein families, which the Henikoffs termed as blocks. These are directly calculated based on local alignments. So these are all based on observables; so, using substitution probabilities, again, the concept of conservation, and the overall frequency of amino acids. Because of the way these were constructed, using known protein families, they're really
19:22
sensitive to detecting structural or functional substitutions, and they perform better than other matrix series that are out there, specifically the PAM Matrices. The PAM Matrices were developed way back in the distant past, as far as bioinformatics goes, at a time where very little sequence information was available; so we only had sequence data available for globular proteins, they were -- the matrices were extrapolated from closely related proteins, and you'll
19:51
often see them as an option for a choice in phylogenetic programs, but I would recommend that you pick these BLOSUM Matrices instead, because just bottom line, they really are useful in identifying both closely and distantly related sequences. So that's a big plus. All right, so now here's where evolution comes into play. When I showed you the BLOSUM Matrix earlier, that was a BLOSUM62 Matrix. So there's always a number that follows the word BLOSUM.
20:20
And what the N represents is the degree of similarity of the sequences that were used to construct the matrix. So in -- so here you've got -- in the case of a BLOSUM62 Matrix that means that the sequences used to construct the matrix shared no more than 62 percent identity. In the construction what happens is that the contribution of sequences that are higher than the threshold are clustered and replaced by a single sequence that represents
20:48
the clusters. So what's the reason for doing this? Last week Eric showed you the phylogenetic range of a number of vertebrate species whose genomes had been sequences, and while we have more and more whole genome sequences in hand from a wide variety of organisms, the genome sequences that we have available to us are not uniformly distributed; we have multiple sequences for some organisms. So we really need to just remove that bias from the data
21:14
set, and that bias tends to be more toward closely related sequences. We want to do this, because we don't want to miss anything that might be biologically relevant. So in practice how does this work? Let's say we start with a protein family. Here we've got one with six sequences in the group. Any place you have an exact match you'll see an asterisk at the top of the column. Let's say we're trying to construct a BLOSUM80 Matrix.
21:40
So we're going to cluster together any of the sequences that share 80 percent or more similarity with any of the other sequences in the set. So in this case we end up with three clusters, one that has sequences, one that has three, and one that just stays by itself; it doesn't cluster with any of the other sequences. We then, now collapse this down, representing these six sequences as three. So these three sequences are now represented
22:06
by this single sequence, these two sequences are represented by this single sequence, and now we use this to construct the matrix. So what are the implications of this? When we do the clustering, again, we reduce this contribution of closely related sequences; so we remove the bias that I mentioned earlier about substitutions that occur in the most closely related members of a family. But here is the more important thing: when you reduce N, the value of N, you're going to get more
22:36
distantly related sequences; if you increase N, you're going to get more closely related sequences. So, here is your first cheat sheet of the morning; put a big star next to this one, and -- as far as which one of these matrices to choose. The default by -- if you're going to do a BLAST search, at NCBI it's going to be BLOSUM62. The reason it's the default is, just through a lot of very controlled trials, it has been
23:00
seen as being the most effective in finding all potential similarities, and you're going to get percent similarities that go down as far as 30 to 40 percent. Okay, as we move up the numbers here, as the number get higher, they're going to be -- we're going to get sequences that are more closely related. So by the time we get to BLOSUM90, we're going to have maybe shorter alignments, but they're going to be much more similar to one another
23:22
in the range of 70 to 90 percent. We go in the opposite direction, we're going to find more distantly related sequences; so longer alignments that might be weaker, and you can push it down to probably around 30 percent. Now, of course the question goes back to which one to choose? If you know what you're looking for in advance, if you want to find things that are highly similar, you're going to push to the higher end of this range; if you want to look for things that are more distantly
23:48
related, you can go to the lower end of this range, but the method that I would strongly recommend you use is something that Steve Altschul, who is a person who developed BLAST, talked about in a paper called -- he called it the True -- "Three Matrix Strategy." And so, instead of doing a single search, instead you're going to do three; you're going to pick 62 as a starting point, pick one that is going to give you more distantly related
24:14
results, one that's going to give you more closely related results. And what you're going to find is you're going to get three different sets of results back. They'll be highly overlapping, but one -- some of them will pop out certain sequences if the other ones won't. So again, I would strongly recommend you use this three matrix strategy as you learn these techniques, as you become more comfortable, more familiar with them, but the bottom line to all of this
24:37
is that no single matrix is the complete answer for all sequence comparisons. All right, so one last thing we need to discuss before we get into talking about BLAST, is GAPS. So again, I want you to imagine Sequence A and Sequence B. And when we do these alignments, often you'll see we'll introduce GAPs in order to improve the alignment, so we get these local alignments better. So again, used to improve the alignments between these two sequences.
25:08
More often than not, these are used to compensate for insertions and deletions, and this is really important, because the gaps -- because we're representing insertions and deletions -- they represent biological events, so you can't just keep putting these gaps in willy-nilly to improve your alignment, because you're not actually reflecting biology by doing that. So you need to keep them to a reasonable number, again, to not reflect a biologically implausible
25:34
scenario. And a good rule of thumb for this is about one gap per 20 residues, and it depends on the case, but that's a good starting point. Now, if you think again about your alignment, think about the scoring matrices we've been comparing, single positions, a letter in one sequence to a letter in another sequence, but now we have a situation where we have a letter in the one sequence and nothing in the other sequence; we have a gap there, so we can't score this simply as a match or a
26:00
mismatch. So how do you score an amino acid against essentially nothing? The method we used to do this is -- involves something called the Affine GAP Penalty, and so what this does -- all you need to know about this is that any time you open a gap, you assess a penalty; anytime you extend the length of a gap, you assess an additional penalty. So this deduction is governed by a formula right here, G+L(N), where G is the gap opening penalty. So as
26:30
soon as you introduce the first position of a gap in a nucleotide alignment, you're going to subtract five points; on a protein alignment you're going to subtract 11 points. As you make that gap longer, you apply the gap extension penalty. So the second term, L(N), so it's the penalty times the length of the gaps. So every additional position you add to that gap in a nucleotide alignment, you're going to subtract an additional two points; in a
26:56
protein alignment, you're going to subtract an additional one point. So let's say we have a gap of length two in a protein alignment. So G here is 11. If the gap is of length two, so L is one times two, the length of the gap. So 11 plus two is 13. You would subtract 13 points for a gap of length two. You can -- as you'll see as we go through the methods, you can actually change the gap penalties to make the introduction
27:25
of gaps either more or less permissive. So if you want things -- if you want fewer gaps, you can adjust for that; if you want to allow additional gaps you can do that as well. All right, so with all of that, half an hour later, by way of introduction we finally get to BLAST. It's finally time to talk about BLAST, and this is really the most important technique in bioinformatics. We're going to hit this pretty hard today, because I want
27:50
to make sure you know how to use it properly. And -- because there, unfortunately, are many cases in the literature where wrong biological conclusions have been made, because of not interpreting BLAST results properly. Either the people doing the analyses didn't know how to use it, or didn't really know what the statistics meant. The most important take-home message I can give you today that applies to this and any of the other methodologies
28:17
that you're going to learn throughout the 14 weeks of this series is that, just because something appears on your screen or on a piece of paper in the results does not automatically imply biological significance. So we're going to talk about what does imply significance in a moment. Okay, so what is BLAST? BLAST: a basic local alignment search tool. What this does is it looks for high-scoring segment pairs, HSP. So again, just a glorified term for just two
28:45
sequences, a pair of sequences, that can be aligned with one another. And when we do that alignment, we're going to try to make that alignment as long as possible in a statistically robust way, and I'll tell you about that in a moment. The score has to break a score threshold, S, before it's reported to you, and the alignments can be either gapped or un-gapped. Again, because we're looking for these local alignments within entire sequences, the results are not
29:12
going to be limited to just the best high-scoring segment pair for the two sequences being aligned, and in fact, I'll show you an example that will produce two high-scoring segment pairs. There are different flavors of BLAST, and three that are used most commonly are the first three on this slide. So BLAST N -- N is for nucleotide. So we start with a nucleotide sequence, and we compare that against a collection of nucleotide sequences. BLAST P -- P is for
29:38
protein. So again, the comparison is of a protein sequence to a database of protein sequences. BLAST X is a little bit different; here we're going to translate the query into a protein sequence before we do the comparison. So, yes, we start with a nucleotide sequence, we do the six-frame translation, we take the protein sequence arising from that, and then compare them against a protein database. So this is actually a methodology that's much
30:05
more sensitive than direct nucleotide searches. The last two are on the slide for sake of completeness. I can't remember the last time I ran one of these. The T in these both stand for translated, and you can see for yourselves what the query and the target sequences actually are, but the comparisons here are all done at the nucleotide level. All right. So how does the method actually work? So we're going to start with a sequence
30:30
of interest. We have an amino acid sequence here, our query. And you'll see I have three letters in the middle of the sequence in red. The P, the Q, and the G. And we're going to start our search with this query word of length three. And what happens in reality is it starts
30:52
at the one end, at the end terminal end, and it just keeps moving along, the window moves along. So it starts with a GSQ, and the SQS, and moves along. But just for sake of symmetry on this slide we're going to start in the middle. What we want to do is now compare this to all of these sequences in a protein database, starting off just looking for other occurrences of the P, the Q, and the G in that order, but we also want to look for conservative
31:18
substitutions, because we might find some interesting biological relationships by doing that. So when we do that we establish something called the neighborhood. So the neighborhood starts off with the exact match, the PQG, and that has a score. So if you were going to go back to your hand outs to the slide that shows the BLOSUM62 matrix, and did -- and find the intersection of the P for the P, the Q for the Q, and the G for the G, you
31:43
would find that the P would give you seven points, the Q would give you five, and the G would give you six for a total of 18. What I've done here on the slide is I've just changed the middle position to show those conservative substitutions. So we've got the PQG, EG, RG, going down the slide. And you'll see that because these matches are not exact at that position, as you would expect, because now you know how the scoring matrices work, the
32:08
score also goes down. And of course, you would do this at all three of the positions. Now, at some point you have to say, well, maybe we're kind of going outside of the neighborhood. So the method will automatically draw for you a line at what it calls the neighborhood score threshold; BLAST computes this for you. In this case, T is 13 given the query word that we started with, so anything above the line, everything from 13 and up, we keep for
32:34
the next step. Everything from 12 and down we throw out. All right. So let's take our neighborhood, put it at the top of the next slide, and here's the sequence now that we have that we started with. Here's our PQG, our query word. You'll notice, now I have it aligned with a sequence that was found, the subject, but in this case it's not a PQG; it's the P, the M, and the G, and that's perfectly fine, because if we look back at the top,
32:58
PMG is in our neighborhood. Now, just qualitatively you can see how good the alignment is by looking at this middle row. So any place you have an exact match you see the letter repeated; any place you have a conservative substitution you'll see a plus sign; any place you see a blank is just a mismatch. All right, but we want to be able to apply numbers to this, and we want to try to make this alignment as long as possible. So here we are, we've got this alignment starting
33:27
at three, but you'll see that there's many amino acids. Here it's -- the alignment is a blank-40. You go as far as you can in both directions, aligning as many of the characters as you can. So the question of course then becomes, how far is too far? Okay. So here's our alignment again. And here's just a graph of how the scoring is working out for this particular alignment. On the X-axis here we have the extension. How many positions have
33:54
we aligned? On the Y-axis is the cumulative score. So the scores that are coming from our scoring matrix as we just keep adding additional positions to this alignment. So if we start out here, the extension is three, what we started with, the query word. We're above the threshold cutoff, as you'll remember from the previous slide. So as we go in both directions we're going to assume that we're having more matches than we do mismatches,
34:20
more positive scores being assessed than negative scores being deducted. So the curve is going to go upwards. At some point, we're going to break a threshold, the score threshold, okay. So this -- from here up, there are the things that are all reported to you in your BLAST result, and herein lies pitfall number one. Breaking the score threshold does not imply biological significance, and I'll show you what does momentarily. We don't pay attention to the scores, but the scores are used to
34:48
calculate the number we are going to pay attention to shortly. Okay. So we're going to keep going up, we're going to keep assuming that we're matching more than we're mismatching, but at some point we're going to have gone too far. We're not going to be able to align any more residues with one another. We start thinking about that matrix again, and all those negative values in the matrix. So as we go out further we're going to start subtracting points faster
35:12
than we are actually adding them. At some point, we're going to leave the top of this curve and go down a certain amount, the significance delay -- decay, excuse me -- that represents both the mismatches, which assess negative points, and the gap penalties, which assess negative points. When you've reached the significance decay, the method says, okay, I've gone far enough; I'm going to go back to the peak of this curve. That's going to represent the
35:38
longest alignment I can put together, the length of my high-scoring segment pair. Okay. So once we've done that we're going to have to think about doing a probability calculation. So again, why do we use the probabilities instead of the scores? Now that you know about the scoring matrices, the answer will hopefully be easier to understand. I want you to imagine two different alignments of equal length. One is composed primarily of common residues.
36:09
One is composed primarily of rare residues. Okay, and the composition of those alignments are going to be reflected in the score. Even if the number of positions in those two alignments that match one another is exactly the same, you're going to end up with two very different scores, because of the frequency information that's captured within the scoring matrices. So a better measure is needed to be able to compare one alignment to the next, any two
36:34
alignments to one another, and that again is where the probability comes in. So this is the equation, the Karlin-Altschul equation that actually governs that calculation. You'll see that in the exponent here is the S -- so that is the normalized score -- and the other factors that go into the calculation of this score. Again, you don't have to do this; the BLAST method will do it for you, but what that value, E, represents is the number of
36:58
high-scoring segment pairs found purely by chance. So this is really the number of false positives that you have -- have come out of your BLAST search. We want that value of E to be as low as possible. Lower values signify higher similarity. Okay. So rule number one. Again, put a big star on this one. So what we're going to look for as far as probability values, if we're comparing two nucleotides sequences to one another,
37:25
we want that E value to be 10-6 or lower. If we're looking at protein sequences, we want that value to be E to the -- 10-1/3 or lower. And again, keep in mind that when we talk about protein similarity searches that means both BLASTP and BLASTX. Remember the BLASTX; we start with the nucleotide sequence, but the comparison's done at the protein levels, so you're going to use the protein statistics here. All right, so let me take you through actually how you're going to do this when you get back
37:55
to your offices and to your labs. So we're going to pretend we're at the computer. You have the screen captures in your handout. You might want to look at the PDFs on your monitors back at your desk, because they'll be a little bit easier to read, but the most important thing to do is practice, and just take some sequences that are of interest to you and the biological questions you're trying to answer, and use the methodologies that
38:14
I'm going to show you today. All right. So we're going to start off at NCBI's homepage, which there is the URL, ncbi.nlm.nih.gov. The most popular things that are on this website are shown in the right sidebar, and there is a direct link to BLAST there. So we're going to go ahead and follow that link, bringing us to the BLAST homepage, or you can just go directly by typing in the full URL. So let me quickly orient you to this page. At
38:41
the top of the page you can just go directly to full genomes, and search each one of those genomes individually. The middle of the page, which is where the action is today, is where all the basic BLAST searches are. And at the bottom are a series of specialized BLAST searches, which we'll talk about momentarily. So we're going to start off doing a protein-based search; so a protein BLAST. The link is right there, and if we go ahead and click that it will take us to the BLAST search page. But
39:12
before we do that we need a sequence that we actually are going to use as the query. So what I put together for all of you is a webpage. Here is the URL that has the query sequences for all of the examples that I'll be showing you both this week and next week. So this is a great way for you to go back to lab and do a little bit of practice on your own, reproducing what we're doing in the lecture this morning. Okay. So we're going
39:35
to take that first sequence from the page I just showed you, and just stick it right into the box. Now, you can use the entire length of that sequence -- that's great; if you are only interested in part of that sequence, you can actually specify the beginning and ending amino acids you want to consider the start position and the stop position. Okay. So the first thing we have to do is pick, what do we want to compare this sequence to?
40:00
What database do we want to look at? So we have a bunch of choices available to us, and I want to talk about one in particular that's called RefSeq. So what RefSeq is, is an effort to provide a single reference sequence for each molecule of the central [unintelligible]. So to have every DNA sequence, every MRNA sequence, and every protein sequence represented once and only once. And the reason for doing this is it really became obvious that we needed
40:30
to have a resource like this when a lot of us were doing these BLAST searches, and you might take a gene of interest, you'd get back maybe two, five, 10, 20 exact matches, and you just don't know by looking at the BLAST results which one is the right one; the canonical entry for that particular gene or for that particular protein. So this is a curated effort that helps address that problem. So by virtue of the fact that each of these sequences are represented once and only once,
40:57
of course, the database is non-redundant. It's constantly updated by curators to reflect the current knowledge of sequence data and biology. And that's a great thing: that these records keep up-to-date with what we know functionally about these nucleic acid sequences or these protein sequences. In the featured tables, the part of these entries that captures these biological attributes, you'll find information about the gene, the gene transcript, or the
41:23
protein. They encompass a wide taxonomic range, but primarily focus on both mammalian and human species; and again, just to reemphasize the ongoing update and curations. And you'll see a time stamp showing you when that record was last brought up-to-date. Now, the way to know what a sequence -- whether a sequence is a RefSeq sequence is by looking
41:49
at its accession number. So anytime I use the term accession number that is a unique identifier that if you type that in it would produce back that one sequence unequivocally. So it's basically that sequence's Social Security Number. And there's two series in the RefSeq accession numbers that I'm going to show you. The first ones, they start with an N. And these come from curation of GenBank entries. So the Ns, let's say we start with an NT,
42:16
that's a genomic contig. It may have come from a whole genome sequencing project or some other high throughput sequencing effort at the nucleotide level. The NM, M stands for MRNAs. The NPs, P stands for proteins. And the Rs stand for the non-coding transcripts. So what these all have in common is that these have all been experimentally verified. Somebody has had these molecules in their hands, and have actually directly sequenced them.
42:42
Now, of course, we crank out sequences at break neck pace nowadays, and so we don't always have the luxury of having these experimentally verified sequences. So what we do is often apply computational techniques to predict where the MRNAs are, and where the proteins are. So this is where this X-series come from. So let's say we start with a genomic contig, these NT entries. And we then take those, apply the predictive methods to figure out
43:12
where the genes are -- so gene structure: introns, exons, and so on -- to figure out where we can come up with model MRNAs, and then we use those in turn to come up with the model proteins, the XPs. So please keep in mind that, again, these are experimentally verified; these come from genome annotation. These are computational predictions. So you have to look at them, not necessarily skeptically, because the predictive methods are so good
43:38
nowadays, but you just have to keep in the back of your mind that they haven't been experimentally verified. So whenever you have a choice you're going to go look at the ones in the N-series instead. One of the other choices that was on the pull-down menu was Swiss-Prot. Similar idea; we'll talk about that in greater detail next week. So I like to use RefSeq, just because -- this non-redundancy, you just get a neater and shorter hit list from your results usually.
44:05
Okay. So let's go back to the screen here. So now, we go a little bit further down. I
44:13
-- what I've done -- excuse me, for the purpose of the example I've just left the database at the default, which is the non-redundant protein sequence set. So that's every sequence that's in GenBank. As far as the organisms go, you can just leave it blank and search for everything, but you can limit by organism or by taxonomic groups. So you can put in the word human, you can put in the word bilaterian, you can pick particular classes if you're
44:36
looking at a specific point in evolutionary time. Okay. As we go down, program selection. So BLASTP is already chosen for us, but you'll see that there's a whole bunch of other variations of BLASTP, and we're going to talk about those in great detail next week. All right. As we get to the bottom of the screen you'll see that there is a link here that says 'algorithm parameters.' And if you click on that it's going to expand the screen,
45:03
and what we're going to see looks like this. So this is where we get fine-tune what BLAST actually does. First one asks us, well, maximum number of target sequences. How many sequences do you want to see in your results? The default is 100. I always hike this up as far as possible. So we're going to put that at 250 to make sure that we don't miss any statistically significant hits. We go down, word size is three. So you already know how that works.
45:31
If you want three to -- if you want to just use the default, that's fine; if you want to push more towards exact matches, you're going to make the word size longer. The E value threshold, right here it says it's 10. Remember that I've told you that we want to find -- use an E value of 10-1/3 for our cut-off. So I'm just go ahead and leave it at that
46:00
value for purposes of the example, but we're going to apply the cut-off when we see the results. Scoring parameters, we can chose our matrices. Here's your choices, again. All of the BLOSUM matrices -- remember what the numbers actually mean -- but you also have a couple of PAM matrices available to you as well. Okay. Finally, when we get to the bottom, a couple of new things to talk about. So filters and masking. So the first one says 'filter low complexity regions.' We're going to go
46:26
check that off. What is a low complexity region? So these are regions of biased composition. So you might have an amino acid sequence or a nucleotide sequence where on the nucleotide side you might have a string of A's, a very long string of A's; or on the protein side, a single amino acid repeated over and over and over again. So these homo-polymeric runs, you might see short period repeats, or the subtle overrepresentation of several residues.
46:51
The reason it's important to mask these is it actually confuses BLAST. BLAST relies on having uniformly distributed amino acid frequencies. So when it hits these homo-polymeric runs, it often leads to false positives, because if you think about the alignments, and you've got these long runs of the same letter over and over and over again, sometimes it's not exactly clear how to actually align that part of the two sequences
47:16
with one another. So not enabled by default; you should just always do that. Okay. Finally, the last thing, just is a matter of convenience, I always check off the box at the bottom that says 'show results in a new window.' So once we click the BLAST button down here at the bottom, either a new tab or a new window will open. Why that's nice to do is if you want to come back and change the parameters, you can just go back to this window and do that,
47:44
rather than having to keep clicking the back key on your browser. All right. So we'll go ahead, submit this. Here's what we get back. So here we've started
47:55
with -- our query sequences is represented in this case by the grey bar here. Right under the grey bar we'll see that there's a block here, and that represents a protein family that was identified; so a particular domain architecture. Again, we're going to talk about that much more next week. You'll recall that the default for how many sequences to return, how many hits to return was 100. In this case, if we look here it says that 225 BLAST hits
48:24
were found. So it's good that we put our value there up to 250, because we just wouldn't have seen all of the results that BLAST could find for us. Okay. Let's push this overview to the top here. So this is a graphic overview of our results. So here our sequence, our query sequence is shown as the red bar, and you see just a bunch of reds and purples here. So what is this mean? All of the colors -- the color key is
48:52
at the top. Remember that the colors correspond to the scores, not to the probabilities. These are shown -- so there's a color key -- these are shown in descending score order; so the best score is at the top, the worst score is at the bottom. If you mouse over any of these lines, there's a window here at the top that will tell you what the found sequence was, what that particular line corresponds to, what sequence it corresponds to. You may
49:19
see these lines in between. So what the gaps will -- these either represent gaps or that you have more than one high-scoring segment pair for the query sequence versus that particular sequence. If there -- and you'll see that either as a gap, or you might just see this tick mark here, which means the gap is actually in the query sequence rather than in the found sequence. Finally, you might see lines that have these bars on them, but there's nothing
49:46
connecting them. Those are just unrelated hits, just done that way to save space. Okay. So let's go ahead to see what we actually got. Let's scroll down, take a look at the hit list, and see what we can learn. So let me orient you to the results first. You have a description of each one of the sequences that was found that was deemed somehow related to our query sequence. Each one of them has a short description called the definition
50:11
line or the def line; just a quick one-line description of what that sequence actually is or does. You'll see that there are a couple of columns here that give us the score values. We have query coverage, so we'll see that in a little bit more detail, but the important column here is the E value. So I sorted this by E value. Next to that, you'll see the percent identity values, and then these are the accession numbers; again, the unique identifiers, the
50:38
Social Security Numbers for each one of these found sequences. All right. So if we look at the very first one, we see the E value is zero. So what the zero tells you is -- just because of space there's not enough room to show the number -- so whenever you see zero it's 10-1000 or less. So you don't even have to think about those. Those are great hits. We look a little bit further down, here we have a case 3e minus 179, so that just means
51:04
3*10-179. So it's just showing the exponential notation there. I'm going to scroll down to the bottom. And remember that this is now a protein-protein search, so the E value cut-off that I'm going to apply is 10-1/3. Now in this particular case -- again, here's the E value column -- this particular case, the one at the bottom of the list actually beats the threshold quite handily: it's 10-53. So we're going to go ahead and just draw the
51:32
line, and accept all of these values for now. If we saw any that were below that 10-1/3 threshold, we would go ahead and reject them, but I want to provide a caution here, because you might have a case of something that's just below the line that actually is biologically relevant. So this is where you have to put your biology hat on -- you've read tons of papers, you've done your own research, you know something about the system that you're
51:58
looking at -- and you may have biological evidence that argues in favor for something being slightly below the line as being related. So anytime you have biological evidence that always, always trumps computational predictions. So the guidelines that I give you here, these cut-offs are starting points, but again, let your knowledge of biology be your guide here. It will never serve you wrong. Okay. So, let's take a look at one of the results that came out. And so here we have
52:29
a -- the classic way of representing a BLAST search, so where we can see the alignments. So at the top we have some scoring information, our -- which we'll come back to in a moment. We can see where this sequence aligns. So if in our query sequence we're starting at position 17, the first line goes from 17 to 76, and you can see the line numbers -- excuse me, the position numbers going all the way through this alignment. So the first thing
52:59
we want to do is take a look at these values that we have calculated to give us yet another way of determining whether or not these are actually related to one another. So first we have a percent identity number. In this case, they're 100 percent identical to one another. The numbers you want to look at -- this is the second criteria, so again, put a star here -- if we're doing a protein-based search we want to see percent identity greater than
53:26
or equal to 25 percent. If we are doing a nucleotide based search we want to see a percent identity greater than or equal to 70 percent. Right next to that you'll see where it says positives. In this case it's still 100 percent, so what the -- here what that is, is really a percent similarity. It includes conservative substitutions. So again, identity means identity. Positive means the identities plus the conservative
53:50
substitutions. In this particular alignment, no gaps, so we don't have to worry about that. But what you will see is that there are some regions of our query sequence that appear a little bit greyed out in lower case letters. Those are the low complexity regions, the regions of biased composition where you have the subtle overrepresentation of a couple of amino acids. Here's a great case where there are -- you can just see over and over
54:15
again just short period repeats, or the same amino acid coming over and over again. Okay. So that is one high-scoring segment pair that came from the comparison of our query sequence to this particular protein, this prospero isoform, but if we scroll down a little bit more we'll see something that says range two. So we started with our two sequences and two different high-scoring segment pairs were found. So there's two different parts of the
54:45
alignments led to these local alignments. So again, remember the difference between global and local alignments. So instead of forcing the alignment across the entire sequence, we're just trying to find those local regions of similarity. In this case there are two. So we have the second high-scoring segment pair detected. We're going to look at our identity information again. So 95 percent identical, the starting position for that
55:10
particular part of the alignment. Here we do have a gap and that's just represented by the dashes, in this case, in the subject sequence. Okay. So let's put all of this together. We go back to this graphic. Here are the two scoring lines from the two alignments that I just showed you. So let's think about our cut-off. So first we want to make sure that E value is 10-1/3 or better. So, yep, we've got that. So zeroes in both cases. We look at the percent identities in protein base,
55:39
so we want that to be 25 percent or better. We got that as well. So 100 percent for the first one, 95 for the second. And if we just look at where these correspond to the overview that we had, the first high-scoring segment pair goes from position 17 to 704; the second one from 777 to 1403 with a small gap between them. So the vast majority of this sequence is align-able, but there's a region in the middle that you can't align the one you started with, with the one that you found. All right.
56:09
So your second cheat sheet for the morning -- again, keep this handy -- if we're talking in terms of E value at the nucleotide level 10-6 or smaller; protein level 10-1/3 or smaller. As far as sequence identities go, for nucleotides 70 percent or more; proteins, 25 percent or more. Again, don't use these cut-offs blindly. I give this to you as a starting point, as you start to learn these techniques, and start to get a better sense for how they work, but
56:41
please, please do pay attention to the alignments that are either side of that dividing line, and never, ever ignore what you know about the biology of those sequences. That always is the most important thing to factor into these analyses. Okay. So let's take a little bit of a simpler approach here. Let's say, instead of searching a database, you have two sequences and you just want to see how those two sequences align with one
57:05
another; their predetermined sequences. So just like a BLAST search, we're going to find these local alignments, either nucleotide sequences or protein sequences. All of the BLAST algorithms are available to you. Again, you can select either BLOSUM or PAM as the matrix for these comparisons. Same Affine gap costs are -- can be applied here. You can adjust them. The input sequences, again, can be masked. So we come back to the BLAST
57:31
homepage. In this case the link to get to this is in the lower section in the specialized BLAST section, 'align two or more sequences using BLAST,' BLAST two sequences. When we click on that link it brings us to this page. It looks very similar to the one we saw before for the BLASTP search. We're going to look at these top -- tabs at the top first, make sure that we've clicked on BLASTP in this case to see the search page that's relevant to a protein-based search.
57:57
If you go to the example sequences page that's up on the current topics website, you'll -- we're going to just take the two sequences from those sample sequences; paste one in the first box, paste the second one in the second box. As before, we can specify query sub-ranges or subject sub-ranges if we don't want to use the entire lengths of these sequences. Once again, at the bottom, the algorithm parameters link; we're going to go ahead and expand the
58:26
page now by clicking on that. We'll now see all of the stuff here at the bottom. So the general parameters here, as before, how many sequences do you want returned? Well, we already know we've got two, so we can just leave that alone. The E values, the word size, as before. Here I actually don't change anything, because you already know that you've sequence A and sequence B, so you're just going to, by inspection, not have to worry about any of these things.
58:53
You're just trying to align the two sequences. But what will come out is the statistical measure of how well those two actually align with one another. As before, you can choose of a series of matrices. We can mask out the low complexity regions, which you should always do by default. Again, we're going to check off the little box at the bottom where we'll show our results in a new window, and then just go ahead and click the big BLAST button at the bottom of the page.
59:21
So what we get back is reminiscent of what we just saw. We have our overview of the results. Here's our query sequence, again, as the red line. We only searched it against one sequence so there is the one sequence. Part of it aligned well, part of it a little less well. Let's scroll down a little bit to see what those results look like. So in our data table up here, again, one and -- only one hit. We see the E value and the percent identity. A greater
59:49
amount of information shown in the lower part here. So here we have our first part of the alignment, the longer part. The percent identity's 46 percent, so that beats our eval -- excuse me, our percent identity threshold for proteins, the expect value; 10-26 beats our 10-1/3 threshold,
1:00:11
and again, you can just see what part of the sequences align. So in this case the query from positions 95 to 178. The second part, that little part at the end terminal end, if we look at the stats here, we're just -- it's a very short alignment; as you can see here, it's only six amino acids long. The E value is almost 2, so we're just going to go ahead and reject that just on the basis of that one number alone. We can also look at this in terms of a dot-matrix;
1:00:36
it is a traditional way of looking at this. Sequence 1 is represented on the x-axis, so positions 1 to 178; Sequence 2 on the y, positions 1 to 134. We saw that there were two high-scoring
1:00:50
segment pairs, or at least two alignments, one of them, which we actually accepted. Here is that shorter alignment, so it just shows you graphically which parts of the sequences were aligned with one another. It's a little bit more intuitive to look at this, but I thought I would show you this just for the sake of completeness. All right, so we've now spent a lot of time talking about things at the protein level,
1:01:10
but of course, there is a lot we can learn by looking at the same types of information at the nucleotide level, so we're going to talk about a couple of methods here. All of the nucleotide BLAST algorithms can be found in the basic BLAST section by clicking on the nucleotide BLAST button. And when you get to the program selection part of the search page, you'll see that there are three options available to you. Mega BLAST, discontiguous Mega BLAST, and BLASTn.
1:01:37
So, what are those doing for you, what do they actually do? First one, Mega BLAST, is the one you would pick if you were searching against complete genomes or chromosomes. So this is really optimized for aligning really, really long or highly similar sequences; things 95 percent similar or more. You'll see the word size, because we're looking for things that are almost exact is very high at 28, the Affine gap -- excuse me, this is the scoring
1:02:07
matrix. So for a match you get one point; for a mismatch you lose two. This is a linear gap penalty, so unlike the Affine gap penalty, just a value is assigned, the same value for opening the gap and for extending the gap. Okay. The middle block here, talks about discontiguous metal -- Mega BLAST, and this is what you'll be doing most of the time at the nucleotide level. The results between these two are very
1:02:35
similar, but Mega BLAST will run faster in relative terms. So here we've got discontiguous Mega BLAST; here is BLASTn, our standard; we've got a nucleotide search that we're comparing against a whole slew of nucleotide sequences, or nucleotide database, and you'll see that the values are all the same here. Finally, in the lower block, if you've got things that are very short, nearly exact matches, again, you can use BLASTn to do it here, but rather
1:03:01
than using the default, you can just change the values here to push in that direction. All right, so enough about BLAST. Let's talk about something else that you could put into your armamentarium, and this is a method called BLAT, which is the BLAST-like alignment tool. This is a method that was developed at the University of California Santa Cruz, and it's very similar to Mega BLAST in that it was designed to rapidly align really long nucleotide
1:03:31
sequences, so sequences at least 40 in length, having a 95 percent sequence similarity or more. It reliably can find exact matches down to a length of 33, but this is really the method of choice when looking for exact matches in nucleotide databases. The other thing that is very impressive about it is its speed; it's 500 times faster than BLAST for most MRNA or DNA searches. The downside: it might miss divergent or shorter sequence alignments,
1:04:01
because it's not optimized to do that; it's optimized to find that these long stretches of similarity. You can use it on protein sequences, but if you're looking at protein sequences you should really be using BLASTP instead. So, when do you use this? You've now done a sequencing effort, you have got a brand new gene that you're trying to characterize, or a sequence fragment that you're trying to get information on, so you'll be able to find its genomic coordinates, determine gene
1:04:29
structure -- so where the introns or extrons might be -- or find markers of interest in the vicinity of that sequence. You can also use it to find highly similar or identical sequences. So if you have a ton of MRNA data, you could align those MRNA sequences onto a genome assembly, you can start to identify additional genome family members, or you can start doing cross-species alignments to identify putative homologs, so going back to the beginning.
1:04:57
The other thing that's nice about it is you can take a sequence and show it as a separate track within the UCSC genome browser. So this is a new concept, just thinking about a track -- we'll see an example of that shortly, and Dr. Wolfsberg will be spending quite a bit of time discussing the use of the UCSC browser when she presents her lecture in two weeks' time. So let's just go ahead and go to the BLAST -- excuse me, the BLAT homepage at genome.UCSC.edu.
1:05:24
The link to BLAT is in the left sidebar, so if we just go ahead and click on that, it brings us to a very simple search page. Across the top you have a small number of selections you can make, so you can pick which genome you want to search against. In this case I am starting with a rat sequence, so we are going to search against a rat genome. The query type here is DNA, and then the other parameters are predefined for you. Then just
1:05:50
go ahead and click on submit, and we're going to get back a very easy to understand results table. Okay. So across the top we have a number of sequences that were found based on our query, you'll find a score for each one of the sequences that were found based on our query. You'll see a score for each one of them; so these are in reverse score order, the highest one at the top; starting and ending positions for each one of those alignments,
1:06:15
and a percent identity. We're going to go ahead and click on the browser link for the very first one, the best hit in the group, just so we can see what this now looks like in the context of the UCSC browser. So we see where our chromosomal position is here by the red bar on this chromosome, on Chromosome 5 of the rat genome. Our sequence is the one that is labeled, quite self-explanatory 'Your Sequence from BLAT Search.' So there is the sequence, and you'll notice some discontinuity
1:06:46
in the end; so it's not a perfect alignment across the entire length. There's some gaps in there, and there are also some regions that are marked in red along the way. So the reds are mismatches. You'll see the color key here at the bottom, so red genome inquiry sequence have different bases at this position, and you can read the rest of them for yourself; the other colors are in the key. What we also see here are other tracks. So each one of these lines is a track, so we
1:07:13
have information on rat MRNAs from GenBank, Rat ESTs, the expressed sequence tags, some conservation information; so looking at data from mouse, and human, and dog, and opossum, and so on. So it's a really powerful way of not just finding what the chromosomal context of your gene of interest is, or sequence of interest is, but a lot of other information about related sequences, other types of data that go well beyond just simple idea of alignment.
1:07:42
And again, Tyra is going to spend a lot of time talking to you about this during her lecture in a couple weeks. All right. If you go back to our first page, we have another set of links here called 'details,' so details link for our best hit. And so here what we have, the top part of this is our query sequence, the one that we started with from our sample set of sequences on the current topics sequence page. This is the sequence that was found. You'll see that there are
1:08:12
things in different colors and upper case and lower case; the key is at the top, so anytime you have a matching base in CDNA and genomics sequences those are blue and capitalized; the things that are in light blue are the boundaries of gaps in either sequences, and those are often splice sites [spelled phonetically]. So, it's a view; it's not a particularly informative view. The view that I think is much easier to look at is the one that is reminiscent
1:08:34
of the BLAST view, where we just have the two sequences aligned with one another, you can see the positions of the gaps along the way, and get a good sense for how aligned these two sequences are with one another. Finally, I just want to talk about one last sequence alignment program, and that's called FASTA, and so what FASTA does is identifies regions of local alignment, must like our
1:09:04
BLAST method does, but it uses a different algorithm to use it; it uses something called a Smith-Waterman Algorithm to find the best alignment between the two sequences. So the methodology that underlies this technique is significantly different than the one that is used for BLAST. Here are the URLs if you'd like to check that out and play with it a little bit. It's slower than BLAST, but it's more stringent in certain ways, so I think
1:09:27
one way to include this in your armamentarium is, if you think back to our BLAST results, we would draw a line at our E value cutoff, you could use it as a second method to rule in or rule out things around the cutoff. So the same thing -- way that in a lab you might use two different experimental techniques to verify a result, here you can use two different computational techniques to verify your computational results.
1:09:51
Okay. So with that, that's where we're going to leave things today. We're going to build upon the basic concepts next week when we start to discuss how we can go well beyond, just looking at letter by letter alignments to find commonality between sequences, and we're going to end up talking about how we can use multiple sequence alignments, and apply those to your own work. So thank you once again for joining us today, and we'll see you again next week. [applause]
Here to help you learn
Sign in to keep a private conversation with your course tutor.
Ask me anything about the material…
Sign in to chat