Friday, August 13, 2004

Immunology is cool

I had a chance to talk to Dr.Luk Van Parijs today, who works on the immune system. His work seemed interesting to me because I'm interested in doing research on topics that are relevant to disease control, and studying the immune system seems like one good way to do so. He's doing a bunch of interesting stuff like

- Using RNAi and lentiviral expression vectors to decrease and increase the expression levels of certain genes in mice and then seeing whether that changes the susceptibility of these mice to tumors. The idea is that, instead of starting by a priori studying the genes that we know are relevant to immune system and then expanding out from there, you can make the assumption that all genes are relevant to the immune system and then figure out which ones aren't by knocking them out or enhancing them and showing that they don't impact the immune system response. The flip side, of course, is that concentrating on whether the genes affect the immune system response to tumors means that they won't detect genes that affect immune system response to, say, the tuberculosis bacillus, but it's basically a "crawl, walk, run" strategy -- focusing on just one system at first allows them to produce pretty crisp results about at least one system, without having to worry about side-effects on other systems. Part of the problem here is how much data they generate and how to actually interpret it -- they basically have a data mining/machine learning challenge.

- Computational modeling of certain pathways in the immune system, like cell death caused by killer T-cells attacking an infected cell. One of the big challenges in setting up these computational models is that getting quantitiative data to base the models on is pretty hard -- you have to question whether you actually believe the data [eg certain labs have the reputation of not producing very good data], you have to deal with the fact that there is no consistent standard for how to store biological data [eg in Excel, in just a plain file, in an Oracle database], you have to comb through the thousands of papers on the topic to figure out whether you have all the major players in the pathway etc. However, the model they've built has been pretty successful for them, and so now they're looking at expanding into other areas and doing things like statistics-based modeling -- if you have lots of different moving parts [eg reaction rates, temperature, concentration, acidity etc], you can't do experiments that cover all the possibilities, so how do you know which experiments to do that will give you the most important aspects of the model ? Lots of interesting computer science/computational-type questions to answer in that area.

The immune system in general is very interesting -- it's made up of many different actors [B-cells, T-cells, dendritic cells, antibodies, pathogens etc] all affecting each other in rather non-deterministic ways, yet doing a pretty good job overall of dealing with the myriad of microbes we're exposed to every day. It's very much a "systems" problem because everything is interconnected, and so it's rife with opportunities to build computational models to see how changes in one area affect other areas. Depending on how things go, I might end up doing one of my rotations in his lab to get a better understanding of what the possibilities here are.


Post a Comment

<< Home