Loss of information at deeper divergences, and what we can do about it – Distinguished Professor David Penny

It has been shown by Mossel and Steel (2004) that simple Markov models lose information at the deepest divergences (say, greater than 400 million years ago); and that the fall‐off is exponential at deeper times. However, that does not mean that there is no information left; for example, the three‐dimensional structure of proteins should still retain information about deeper divergences, although we may not yet know how to use that information. Biologists still want to estimate the deeper divergences and thus it is a significant question to find additional sources of information. Several suggestions are offered that require a more formal analysis. Firstly, we probably expect that where there is a real Gamma distribution of rates, information may be retained for longer. Secondly, if there is really a bimodal distribution of rates, then identifying, and eliminating these faster‐evolving sites should help. Thirdly, the inference of ancestral sequences at deeper divergences appears quite robust, and there is some evidence that this may help recover deeper divergences. Fourthly, it is increasingly possible to infer three‐dimensional structures, and these should retain information longer. Fifthly, there may be differences between the loop regions of Akaryote and Eukaryote proteins, and only taking the regions crossing the central 3D region might help. Sixthly, an approach of weighting, not of characters, but of the partitions they are consistent with, might help. Seventhly, possibly gene order information might be helpful. Several examples of such approaches will be presented, and a challenge issued to theoreticians to solve some of these fundamental issues. There is still a lot to learn about protein evolution.


Comments are closed.