Let’s look at Sanford’s numbers. A key bit of his paper is this: “We model 10% of all mutations as being perfectly neutral, with the remainder of mutations being 99% deleterious and 1% beneficial [35] […] We use the well-accepted Weibull distribution for mutation effects (a natural, exponential-type distribution [26]). In this type of distribution, low-impact mutations are much more abundant than high-impact mutations.”
Reference 35 is to this paper. In that paper, the authors report on 91 synthetic mutants of an RNA virus. They found 24 produced no virus and can be presumed lethal. If we ignore those, 31 (46%) had no statistically significant effect on fitness, 32 (48%) were deleterious, and 4 (6%) were beneficial. If we assume (without justification) that all mutants with lower measured fitness were deleterious, even if the value was not statistically significant, we have 76% deleterious and 18% neutral. They also found that the distribution of fitness effects detectably non-lethal deleterious mutations had a longer tail – more highly deleterious mutations – than could be fit well with an exponential-type distribution. Mean beneficial effect was 1%, mean/median deleterious effect (for non-lethals) was -13.9%/-9.2%
So let’s compare the Sanford model with the empirical results from the source they cite. Fraction neutral: 10% (model) vs 18-34% (empirical); fraction beneficial: 0.9% (model) vs 4-6% (empirical); fraction deleterious: 89% (model) vs 46-76% (empirical); size of beneficial effect: maximum of 1% (model) vs mean of 1% (empirical). Sanford’s paper doesn’t give the mean fitness effect of their deleterious mutations, but does report that 10% had an effect > 10%. The empirical median value for deleterious mutations (assuming all of the non-significant ones were actually deleterious) is just under 10% (9.2%), which means that Sanford’s distribution is weighted far more toward mutations of small effect than the empirical values, i.e. weighted toward mutations that can accumulate rather than be purged quickly by selection.
In summary, Sanford and colleagues took (and cited) a study of empirical values for the very thing they’re modeling, and then discarded every single one of the values and replaced it with one that suited their thesis. I’m not going to bother looking at the rest of their model, and I think I’ll skip characterizing the quality of this effort so I don’t get banned from PS.