Given a bunch of DNA sequences and some pre-processing, LVB gives you an idea how they are related. If you give it one sequence per species, it gives you an idea of how the species are related. LVB combines an optimality criterion that's rapid to evaluate and a subtle heuristic search - with the intention of working fast, and reasonably well, with large input.nick.mccloud wrote:Can you tell us what it is/does.
And type slowly, we're not bad at computers but can't vouch for us keeping up with anything else!
Yes.So it's a way of deriving a phylogenetic tree from a bunch of DNA sequences (aka a ' a nucleotide multiple alignment')?
A hypothetical phylogenetic tree is evaluated by some quantity. For parsimony, this quantity is the smallest number of mutations required, for the particular tree to lead to the particular sequences we have. This is known as the 'length' of the tree. For a particular tree, exact length is easily discovered e.g. by the Fitch algorithm. One can do this by hand for small cases and it's certainly no problem for a computer.& simulated annealing is a kind of local search that helps figure out how close pairs of sequences are by making small random changes to them (much like genetic mutations) until one observed sequence changes into another?
I read online somewhere that the Raspberry Pi does for computers what biros (or Bíró?) did for pens. I agree!As a matter of interest, why did you do a raspberry pi version?
That's also a nice clear explanation - thanks.rew wrote:What, when properly implemented, works a LOT better is simulated annealing. Instead of rejecting ALL solutions that are worse, and accepting ALL solutions that are better, you make it a random process: worse solutions are accepted with a smaller chance than better solutions. In the beginning you start out with the chances reasonably equal. It doesn't make much difference if a solution is worse or better. But as the optimisation continues, you start changing the chances. The further you go along, the chances of a better solution to be accepted get better. And the chances of a worse solution to be accepted get worse. In the end you get a 100/0 chances distribution and you are left with a standard "hill climbing" step.