Re: [Denovoassembler-devel] Ray Cloud Browser - Feedbacks

[Sébastien B. <seb...@ul...>]

On 11/09/2012 11:50 AM, Charles Joly wrote:
> You said:
> "
>      TATCG -> ATCGA -> TCGAC
>      AATCG ->       -> TCGAC
>   I don't see how you can represent this with just one letter."
>
> I'm not sure I get your example. You seem to have 2 identical k-mers

It's a typo. Should be:

       TATCG -> ATCGA -> TCGAC
       AATCG ->       -> TCGAG


> (TCGAC) that are linked with the same k-mer (ATCGA):
>
> TATCG---
>                |---ATCGA---TCGAC
> AATCG---
>
> Here we have two sequences, right?
>
> TATCGAC and AATCGAC(?)
>
> If 2 k-mers are linked isn't it by definition because they only have a
> single base pair that differ?
>
> Seems to me that every k-mer except those at the right end of the
> graph can be represented by the leftmost nucleotide (since the only
> information about a k-mer that is not present in the next k-mer is the
> first nucleotide):
>
> T---
>       |---A---TCGAC
> A---
>

Yes, you need to bootstrap from one full-sequence k-mer, then you can just add
1 letter.

> For the rightmost k-mers, it's a bit more complicated because there
> are multiple cases possible.
>

Hence one letter is not enough for any k-mer with more than 1 parent or more
than 1 child.

Otherwise, people would be using the 1-letter-per-kmer to store de de Bruijn graph,
right ?


If I take my (corrected) example:


       TATCG -> ATCGA -> TCGAC
       AATCG ->       -> TCGAG


Taking only the last is useless:

G  -> A  -> C
G        -> G

Taking only the first is useless:

T  ->  A  -> C
A  ->     -> G


Why not, I will display all k-mer with only their last letter.
Full sequence will be shown on-mouse-over.

As you said, more stuff will be packed on the screen with smaller objects.

Mathematically, you can spell DNA from a de Bruijn graph by taking the
ith letter of each k-mer, in order, in a path. You then have to deal with
ends. If you take the last, the head must be fully-taken and only the last
is required in tail.
If you take the first, just swap definitions of first and last.

Anything in between is a linear variation.

Thanks for feedbacks.

> 2012/11/9 Sébastien Boisvert <seb...@ul...>:
>> On 11/09/2012 10:02 AM, Charles Joly wrote:
>>>
>>> Salut Seb,
>>>
>>> J'ai regardé ton démo pour le browser avec Fred ce matin. C'est très cool!
>>>
>>> Est-ce que ce serait très compliqué de ne représenter que le
>>> nucléotide qui est unique à chaque k-mer plutôt que la longeur totale?
>>> Ça permettrait de représenter un plus grand nombre de k-mer dans une
>>> fenêtre.
>>
>>
>> Does not make sense.
>>
>> Example of a knot:
>>
>>      TATCG -> ATCGA -> TCGAC
>>      AATCG ->       -> TCGAC
>>
>> I don't see how you can represent this with just one letter.
>>
>> However, for chains  x1 -> x2 -> x3 -> x4, I can do that:
>>
>>      TATCG -> ATCGA -> TCGAC -> CGACA -> GACAC
>>
>> Can become:
>>
>>      TATCG -> A -> C -> A -> GACAC
>>
>> In fact, we can only display any k-mer with a single nucleotide without
>> loss of information for those with 1 child and 1 parent (1-1 k-mers).
>>
>> Lukily, most of k-mers are like this !
>>
>> So what'll do is:
>>
>> - render 1-1 k-mers with the last nucleotide
>> - render other k-mers with full sequence
>> -when the user moves its mouse over a k-mer, it will be rendered in
>> full-sequence mode
>> regardless if it is a 1-1.
>>
>>
>>>
>>> De plus, tu pourrais avoir une couleur par nucléotide et ça pourrait
>>> aider à visualiser plus facilement.
>>
>>
>> Sure.
>>
>>
>>>
>>> Fred a même proposé de ne pas dessiner ni les cercle et ni arête et de
>>> ne représenter qu'une lettre pour chaque nucléotide qu'on pourrait
>>> déplacer de la même manière.
>>>
>>
>> No. It's a graph, we need relationships.
>>
>>>
>>> Charles.
>>>
>>