why aren't they using mutation pictures

[gsgs]

why aren't they using mutation pictures

typically you get a set of m organisms with similar RNA or DNA of length n1
and you want to determine the relationship, the mutations.
The p1 positions where all (or all but c) are the same are not interesting
and can be omitted. Let n=n1-p1.
Then you choose one reference sequence, the common ancestor, typically
the average of the sequences.
Then you make the binary m*n mutation-matrix whose rows are the
sequences,organisms , whose columns are the positions and there
is a 1 ar position (i,j) iff organism i is different from
the reference at position j.
Then you draw the mutation picture where 1s are black pixels and 0 are white
pixels like this:
Then the rows and columns can be re-ordered to show the best grouping
of the pixels into connected areas. Related organisms and mutations are
placed next to each other.

http://magictour.free.fr/seq/mitogi3.GIF

This is straightforward, isn't it ? How else could it be done.How to visualize
the evolution of that set.
Yet I never saw it.I don't even know how others do call this "mutation picture"
Instead people are doing "phylo"-trees to assign the organisms to groups.
with lines and sequence names in it. But this takes more space, is not
rectangular and doesn't so well visualize the distances between the groups.

What's the reason that we don't see the mutation pictures in papers
or webpages ? Is it the way how our science is organized with grants,papers
and peer-reviews and there is just no money to be made from it ?

I don't understand.
(I'm mainly doing influenza sequences)

-------------------------------------------------

how to name it, how to find it , have you better suggestions ?
what are suitable keywords to find it
in fact a google image search for "mutation picture" matrix positions sequences
gave this as top hit, my own previous post

http://seqanswers.com/forums/showthread.php?t=25554

[aaronh]

I like your image for small datasets but don't think it is very useful for large scale analysis. People did just show the segregating sites in old papers when they had only one gene to sequence. This kind of representation is good for one or a few genes but the number of sites/pixel will hide more than reveal once you get to 10s of kb. Plus the question most people really want to answer is the relationship between the sequences and phylogenetic clustering is the best way to get that answer.

[gsgs]

I've been using it a lot for influenza mutations, sets of some thousand sequences,
only listing positions with more than 1 sequence mutating there.

Over phylo trees it has the advantage that you can quickly detect reassortments,
recombinations, that the size is smaller, that you can see how/where
distant groups are related,that you can see in what regions the mutations cluster
(no sorting of columns for that purpose)

For human DNA that recombines a lot you can see the recombination frequency
and mixing of haplogroups in the regions.

[aaronh]

Sure for a exploration of the data, it could be useful. But most of the time, a raw view is not going to make your major point and it is still limited to a max number of segregating sites. The only place I've seen it used well is in "On the Origin and Spread of an Adaptive Allele in Deer Mice".

[gsgs]

when I have many sequences, I just pick 1000 or such at random,
order them horizontally (nucleotide-positions) and vertically (sequences)
to get the black 2d-"regions", which show the clusterings
and evolution better (IMO) than the phylo-trees

[gringer]

We've got around 800 mitochondrial DNA sequences, and I've generated plots like this. Unfortunately, they never look good.

There are patches of the genome that are hypervariable, and then long stretches of conserved sequences. You can treat each variant as the same length in the mutation plot, but then that overemphasises the hypervariable regions and doesn't give you a good idea of the genomic context. You can put a single mutation at the precise variant points, but then that makes it difficult to see rare variants in the middle of a conserved region. You can expand out the variant blocks to occupy the entirety of the sequence up to half-way to the next variant, but that overemphasises the variant-poor regions and hides the hypervariable regions.

Then there's the issue of clustering. Without proper clustering the mutation plots are a jumbled mess. Unfortunately for most cases, it's impossible to assign a perfect linear order to genotyped individuals based on variant similarity, and a lot of time can be wasted trying to get it looking a little bit better than what an automated method can do.

[gsgs]

hmm, to improve it, just "multiply" every mutation by the hypervariability of the region.
Create new artificial nucleotide positions that may get new artificial mutations to extend the
nonvariable ones as compared to the variable ones. Use random numbers to decide this,
so to reduce the total number of locations.
For sorting (=clustering ?) (horizontally and vertically) I use a "traveling salesman" algorithm
to minimize the sum of the neighbor-distances. So far that worked well for me, i.e. with influenza.

You could sort the sequences in 3 or more dimensions and then show 2d-projection-pictures (binary quaders) rotating on keys or mouse-clicks in any direction ....
I haven't tried that yet

---edit-----------
or just do it sepately for the regions, k pictures for the k regions in the k-th
hypervariability-group