Lecture 05

 

 

(Previous lecture) ... (Next lecture)

Homology - I: the Principles

What you should take home from this part of the course

• Understand what homology means and what can be deduced from the fact that two sequences are homologues.
• Understand that the similarity of amino acids' biological functions depends on the amino acid but also on the context.
• Know that sequence similarity can be measured, based on amino acid pair scores.
• Be familiar with the concept of a scoring matrix and with scoring matrices in common use for sequence alignment; appreciate the differences between these matrices and the models embodied in them.
• Be familiar with the principle of a dotplot and tools to generate a dotplot.

 

Links summary

• The Dotlet applet on the Web.

 

Exercises

• Read Walter Fitch's review (pdf) (2000, Trends Genet 16:227-231) on the topic of homology and familiarize yourself with the concepts!
• Read Sean Eddy's summary (2004, Nat Biotechnol. 8:1035-1036) of the theory behind and the development of the BLOSUM62 matrix.

   

Lecture Slides

Slide 001

Lecture 05, Slide 001
It is commonplace today that far-reaching conclusions about biological function are drawn from the inference of homology between proteins or protein domains.

The concept of homology

Slide 002

Lecture 05, Slide 002
The concept of homologous sequences is extensively applied wherever genes or organisms are compared. It is especially important for the study of biology through Model organisms, where inferences about one species are made from data gathered in another. These inferences are justified, because (and to the degree that) the species are related ...

Slide 003

Lecture 05, Slide 003
The word Homology has a precise meaning (in biology) and should not be used differently! Read Walter Fitch's excellent review (2000, Trends Genet 16:227-231) on the topic.
 
We conjecture that paralogous genes have similar but importantly different functions because otherwise one of the copies would be superfluous ,and would be lost under evolutionary drift. Note that we are especially interested in orthologous relationships, because we infer that two orthologues would have maintained the "same" functions. However please note that when one or both organisms have undergone additional duplications after speciation, multiple orthologues may exist. Fitch has proposed the term isoorthologues for that pair of genes with the "same" function among a family. While this is an excellent proposal that would address much confusion, unfortunately the term has not caught on. Thus, rather than rely on a shared understanding of expansive vocabulary in the community, define your concepts explicitly and adhere to your definitions.

Slide 004

Lecture 05, Slide 004

Slide 005

Lecture 05, Slide 005
The equivalence principle might not be obvious. Given that the cenancestral sequences (the last common ancestor sequences) of the pairs {A,B} and {B,C} are different, how do those relationships tell us something about the relationship between {A,C}?

Slide 006

Lecture 05, Slide 006
The answer is based on the fact that there are two, but only two, possibilities for the relationship between {A,B} and {B,C}, drawn above (assuming a simple, dichotomous branching relation between the genes). (1): the divergence between {B,C} happened before {A,B}, at node y. (2): the divergence between {B,C} happened after {A,B}, at node z. Thus we do not know which of the cases is the correct one, the ancestral gene for all three genes can be either x or y ...

Slide 007

Lecture 05, Slide 007
... but we do know that some common ancestral gene exists for all three sequences, and therefore they must all three be homologous.

Slide 008

Lecture 05, Slide 008

Slide 009

Lecture 05, Slide 009

Slide 010

Lecture 05, Slide 010

Inferring homology

Slide 011

Lecture 05, Slide 011
Many obviously homologous genes have very low sequence identity. In this example, the aligned sequences of green- and red- fluorescent protein share 57 of 239 identical residues, i.e. the sequence identity is 23.8%.

Slide 012

Lecture 05, Slide 012
Obviously, the fraction of identical residues depends on the alignment and we need to consider how the right alignment can be obtained. But even before we can start aligning, we need to define a metric between amino acids, to quantify amino acid similarity because the right alignment should give us good similarity, not just a large percentage of identical residues. So the second issue comes before the first issue. And there is an additional second issue: how do we treat sequence insertions resp. deletions in the alignment?

Slide 013

Lecture 05, Slide 013

Measuring amino acid similarity

Slide 014

Lecture 05, Slide 014
Biophysical amino acid properties can be used to group amino acids into sets. Each memebr of this set could be considered to be similar, according to that property. Alternatively, biophysical properties can be tabulated and similarity computed according to a scale. Such numerical scales may include properties as *the free energy of transfer from water to octanol;
*the pKa of the sidechain;
*the volume;
*the accessible surface area (ASA), and many more.
 
For more details, see the Wikipedia article on the standard amino acids.

Slide 015

Lecture 05, Slide 015
In a sequence, all information about the molecular properties of an amino acid is condensed into into the single letter code (or other labels for "abstractions" of amino acids). However, how similar one amino acid is to other amino acids depends on the role it plays for the structure and/or function of a protein. Depending on this role, an amino acid like tyrosine would be considered e.g. similar to other hydrophobic amino acids, or other side chains that can accept H-bonds. However these two groups are non-overlapping sets!

Slide 016

Lecture 05, Slide 016

Slide 017

Lecture 05, Slide 017

Scoring matrices

Slide 018

Lecture 05, Slide 018
A scoring matrix is a computational tool that associates each pair of residues with a number. In this example, the score of the pair {E,D} from an alignment is read out from a scoring matrix.

Slide 019

Lecture 05, Slide 019
The Identity Matrix is valid only at small evolutionary distances (where all similarity matrices give comparable results). Currently, it is primarily used for nucleotide sequence comparisons, where the concept of similarity does not really apply. But the more realistic the model of the evolutionary process is, the less information is discarded. Better models make less assumptions. Many interesting biological relationships have been uncovered precisely because we have been able to perform very sensitive homology searches.

Slide 020

Lecture 05, Slide 020
There is more to this matrix than might seem apparent. Mechanistically speaking, similar codons arise from single nucleotide changes. But functionally speaking, the genetic code minimizes the biophysical effect of mutations! Thus similar codons code for "similar" amino acids.

Slide 021

Lecture 05, Slide 021
The model that M.O. Dayhoff proposed in 1978 departs from an ab initio attempt to define amino acid similarity to an empirical approach.

Slide 022

Lecture 05, Slide 022

Slide 023

Lecture 05, Slide 023
The matrix above is not the Dayhoff matrix, it is just called "MDM". It was derived from the original 1978 Dayhoff matrix for PAM250 by rescaling it to give a constant score for identities (1.5). This may save computational resources but is an unfounded, arbitrary change. Nevertheless, this matrix was the one that was in most common use for many years.

Slide 024

Lecture 05, Slide 024
Two of arginine's six codons (CGG and AGG) can be changed to the tryptophan codon TGG by a single point mutation. Thus these two amino acids, which have quite different biophysical properties and distributions, are defined as almost as similar as a pair of identitcal amino acids by this matrix that was constructed by extrapolation from highly related sequences. The source data that M.O. Dayhoff had used was biased towards sequences in which secondary, functional selection may not have had time to occur after an initial mutation event. Accordingly, this process of extrapolation is expected to inappropriately favour exchanges that can be coded by a single nucleotide substitution. Thus, while the model is rigorous and well designed, the source data limits its accuracy.
 
The inset picture shows the relationship between PAM - percent accepted mutation and % residue identity. This is not a linear relationship, due to back-mutation. In the limit of infinitely many accepted mutations, sequence identity should still be around 5% and not 0.

Slide 025

Lecture 05, Slide 025

Slide 026

Lecture 05, Slide 026
Read Sean Eddy's excellent summary of the theory behind and the development of the BLOSUM62 matrix.

Slide 027

Lecture 05, Slide 027

Slide 028

(deleted)

Sequence comparison: Dotplots

Slide 029

Lecture 05, Slide 029
For examples of biological sequence features in a dotplot: see [http://www.isrec.isb-sib.ch/java/dotlet/dotlet_examples.html here].

Slide 030

Lecture 05, Slide 030
The [http://www.isrec.isb-sib.ch/java/dotlet/Dotlet.html Dotlet applet] on the Web. A of standalone tool for local installation is Dotter, also available on the Web as JDotter.

   

---

(Previous lecture) ... (Next lecture)