Over the past year we’ve started in on a completely new research direction, which concerns B cells, the cells that make antibodies. Antibodies have to be able to bind a very large number of different molecules in order to neutralize pathogens or to mark them for destruction. As part of improving their binding for a given antigen they go through an evolutionary process of mutation (yes, the human genome mutating itself on purpose!) and then selection, whereby B cells producing antibodies that bind well to antigen but not well to self are allowed to multiply. It’s now possible to sequence the DNA or the RNA that determines the shape and binding properties of these antibodies in high throughput. In short, B cell “affinity maturation” is a molecular evolutionary system that evolves very quickly, and one for which data has been quickly accumulating.

Well, we have some experience with analyzing sequence data coming from evolutionary systems, and so we were pretty excited when Harlan Robins (FHCRC PI and founder of Adaptive Biotechnologies) set us loose on a data set he recently sequenced. A natural question to ask is what statistical evolutionary models would be appropriate for this sort of sequence data, and to ask if we can apply some of the tools that we like applying to other evolutionary systems. In our most recent preprint on arXiv we do exactly that. First we do statistical model selection to see how different parts of our sequenced region evolve, and also sorted out a way to do derive a per-residue selection map for this sort of sequence data. The challenge was coming up with a method of estimating selection pressure that would scale to 15 million sequences and handle uneven sequencing coverage. We ended up doing a variant of renaissance counting. Connor and I had a blast working on this with Vladimir Minin and Trevor Bedford.

This is the first (well, technically second) step for our group in this area, and we look forward to continuing.