Getting the US variation manuscript off my hands

Yesterday I gave up and sent the current version of our manuscript about uptake sequence variation off to my coauthors. I'm afraid I'd gotten caught up in generating more and more data, trying to fill gaps in the analysis, when what I really should have been doing was writing a Discussion. But the coauthors will do that, giving me a Discussion draft I can work on further. I'm hoping that they'll also come up with answers and suggestions for all (or at least most) of the issues I've raised in the other parts of the manuscript.

Below I'm going to try to summarize the new data (new simulation runs) I've generated. Right now I can't even remember what the runs were for, and I haven't properly analyzed any of them.

A. One pair of runs were two runs with 10 kb genomes that were intended to split the load of a 20 kb genome run that had stalled (needed only as one datapoint on a graph). That run had used a very low mutation rate and I was trying to run it for a million cycles, but it had stalled after 1.87x10^5 cycles. Well, it kept running, but not posting any more data so eventually I aborted it. Splitting it into two 10 kb runds didn't help - both hung after 1.87 x 10^5 cycles. Now I've made two changes. First, I've modified the 'PRINT' commands so that updates to the output file won't be stuck in the cluster's buffer; this may be why updates to the output files were so infrequent (sometimes not for weeks!). Second, I've set these runs to go for only 150,000 cycles and to report the genome sequences when they finish. This will let me use their output sequences as inputs for new runs.

B. Another pair of runs were duplicates of previous runs used to illustrate the equilibrium. One run started with a random-sequence genome and got better, the other started with a genome seeded with many perfect uptake sequences and got worse. They converge on the same final score, as seen in the figure below.

C. And one run was to correct a mistake I'd made in a 5000 cycle run that used the Neisseria meningitidis DUS matrix to specify its uptake bias. I should have set the mutation parameters and the random sequence it started with to have a base composition of 0.51 G+C, but absentmindedly used the H. influenzae value of 0.38. I needed the sequence that this run would produce, because I wanted to use the sequence outputs of it and its H. influenzae USS matrix equivalent as inputs for another 5000 cycles of evolution. I got the sequence from the first run, and started the second pair of runs, but unfortunately the computer cluster I'm using suffered a hiccup and those runs aborted. So I'll queue them again right now. (Pause while I re-queue them...)

D. Then there were four runs that used tiny fragments - enough 50, 25 and 10 bp fragments to cover 50% of the 200 kb genome. Because the length of the recombining fragments sets the minimum spacing of uptake sequences in equilibrium genomes, we expect runs using shorter fragments to give higher scores. But because the fragment mutation rate is 100-fold higher than the genomic rate in our simulations, most of the unselected mutations in our simulated genomes come in by recombination, in the sequences flanking uptake sequences. This means that genomes that recombine 10 bp fragments get few mutations outside of their uptake sequences, so I also ran the 10 bp simulation with a 10-fold higher mutation rate. These runs haven't finished yet - in fact, most of them have hardly begun after 24 hrs. I think I'd better set up new versions that use the bias-reduction feature, and then run the outputs of these in runs with unrelenting bias. (Pause again...)

The rest of the new runs were to fill in an important gap in what we'd done. The last paragraph of the Introduction promised that we would find out what conditions were necessary for molecular drive to cause accumulation of uptake sequences. But we hadn't really done that - i.e. we hadn't made an effort to pin down conditions where uptake sequences don't accumulate. Instead we'd just illustrated all the conditions where they do.

E. So one series of runs tested the effects of using lower recombination cutoffs (used with the additive versions of the matrix) when the matrix was relatively weak. I had data showing that uptake sequences didn't accumulate if the cutoff was less than 0.5, but only for the strong version of the matrix. Now I know that the same is true for the weak version.

F. Another series tested very small amounts of recombination. The lowest recombination I'd tested in the runs I had already done was 0.5% of the genome recombined each cycle, which seemed like a sensible limit as this is only one 100 bp fragment in a 20 kb genome. But this still gave substantial accumulation of uptake sequences, so now I've tested one 100 bp fragment in 50 kb, 100 kb and 200 kb genomes. I was initially surprised that the scores weren't lower, but then remembered that these scores were for the whole genome, and needed to be corrected for the longer lengths. And now I've also remembered that these analyses need to started with seeded sequences as well as random sequences, because this is the rigorous way we're identifying equilibria. (Another pause while I set up these runs and queue them...)

G. The final set of runs looked at what happens when a single large fragment (2, 5 or 10 kb) recombines into a 200 kb genome each cycle. Because there would otherwise be little mutation at positions away from uptake sequences, these runs also had a 10-fold elevated genomic mutation rate. The output genome sequences do have more uptake sequences than the initial random sequences, but the effect is quite small, and the scores for these runs were not significantly different than those for the runs described in the paragraph above, where the fragments were only 100 bp. This is expected (not that I think it through) because the only difference between the runs is that this set's fragments bring in 2-10 kb of random mutations in the DNA flanking the uptake sequence.

(I was going to add some more figures, but...)