Older blog entries for welisc (starting at number 4)

Surprised to find myself promoted to Journeyer, courtesy of adubey and jlf. I'm pretty good with C and Perl but I'm an Apprentice at best; I certainly don't belong in the same category as people like ewan. Feel guilty now; need to earn my blue. If I had time and had some actual experience with BioPerl I'd offer to write documentation. I'd like to turn my Perl code for handling PDB files into something more extensive (and better documented) along the lines of Konrad Hinsen's MMTK project in Python. The BioPerl folk still haven't added a structural module but most of them seem more interested in sequences at the moment.

Is there going to a schism in computational biology between sequence/genomics people using Perl and structure/computational chemistry people using Python? Interestingly, the survey carried out by the Bioinformatics Template Library developers found that people doing purely structural analysis didn't use Perl at all and Python very little (5% of respondents said they used it); the most commonly used language was Fortran (53%), followed by C (37%) and C++ (5%). By contrast, among the sequence people, Perl was most popular (30%), followed by C (21%). These figures are from 1997 so I don't know how things have changed. I suspect Fortran will decline slowly and Perl and Python will grow.

Just noticed in emk's diary that he describes Java as "sleek". Eiffel is sleek; Java is not, not to my eyes at least, though it's a big improvement on C++ . Each to his own taste I suppose.

Java hbond code running. It's slow, noticeably slower than the Fortran version, which actually does more (calculates bond angles and energies in addition to distances). I suppose part of the problem is that the JVM has to fire up; this is a big overhead for a program that only takes a couple of seconds to run. The IBM 1.3 JVM is supposed to have a shorter startup time but I can't get it to run - a problem in the glibc library. It'd be interesting to see if gcj could improve matters. I want to do a Perl version as well and see how it performs if you push the heavy number-crunching down into C using a SWIG interface. Thought about doing it in Ruby but inertia favours Perl; I already have the code for handling PDB files.

I hate fitting PDF curves. Spent the whole afternoon trying to fit a Gaussian using PDL and it keeps bombing. Tempted to fit the damn thing by hand since I figure it doesn't have to precise. Should maybe try two Gaussians but it looks like a single Gaussian to me.

Downloaded jikes. Pretty zippy.

Day spent fiddling with propensity table code.

Nothing's working. Too little weight on the RAPDF and it has almost no effect. Too much and the stem RMSDs are ludicrous.