Mailund on the Internet

Okay, yeah, so I broke my iMac today, but there is also good news.  We just got another paper accepted, this time a conference paper at this year’s WABI.

Since it is on neighbour-joining, we weren’t that optimistic.  We’ve had problems publishing on this before, but this time it was very well received.

Accelerated neighbour-joining

M. Simonsen, T. Mailund and C.N.S. Pedersen

Abstract

 The neighbour-joining method reconstructs phylogenies by iteratively joining pairs of nodes until a single node remains. The criteria for which pair of nodes to merge is based on both the distance between the pair and the average distance to the rest of the nodes. In this paper, we present a new search strategy for the optimisation criteria used for selecting the next pair to merge and we show empirically that the new search strategy is superior to other state-of-the-art neighbour- joining implementations.

It’s really Martin SImonsen’s work. He is a Mater’s student at BiRC and in one of our algorithmics courses the students were asked to implement neighbour-joining and try to speed it up.  Usually, they come up with some clever ideas, but they never before managed to beat my own version, QuickJoin.

Martin did come up with a faster approach.  Well, pretty close to, anyway.  With mine and Christian Storm’s help, we managed to fix a few things here and there, and speed his approach up to one that not only beats QuickJoin but also all other methods we could get our hands on.

QuickJoin uses a lot of tricks to speed up the search for nodes to join in the algorithm, but the data structures makes it slow on small data sets and also rather memory hungry.  Martin’s approach is much simpler and this helps it a lot in the small data sets and doesn’t seem to hurt it on the larger data sets.

As for QuickJoin, the trick is to only look at pairs of nodes that can potentially be joined and avoid looking at nodes that we can rule out as the next pair to be joined.

Instead of using quad-trees and various functions to rule out pairs, Simon simply sorts nodes in a way where most likely pairs are considered first, and such that we can recognize when new pairs will not be better than those we have already seen.  Read the actual paper — it is quite easy to understand the algorithm from there — if you want the details.

---

Simonsen, M., Mailund, T., Pedersen, C.N. Accelerated neighbour-joining. Proceedings of WABI 2008

My new iMac died on me today.  It crashed unexpectedly three times — and it has never crashed on me before — and after the last crash it didn’t come back up.  I cannot even turn it on anymore, so I guess it is a hardware problem.

I guess the warranty is still good — I bought it this year and there should be one year as I recall — so I’ll send it back to have it fixed.  It still annoys the hell out of me, ’cause I was just getting used to using a Mac instead of Linux.

Oh well, at least I get to test if my Time Machine  backup is working.

First they tell us that alcohol hurts our publication record, and now it causes cancer?  Stop researching alcohol!