[continued from previous message]   
      
   Population genetics is limited by the mathematical and simulation   
   analysis that they have been able to do.  A lot was accomplished before   
   we had computers to play with, but even with computers population   
   genetics is still limited due to what is not understood about the   
   biology.  It is still subject to GIGO (garbage in garbage out).   
      
   The basic underlying biology of qualitative genetics (Mendelian   
   inheritance) was worked out within a century of Mendel's mathematical   
   approach.  Chromosomes were discovered to behave in Mendelian fashion in   
   meiosis (the production of sperm and egg) and we figured out what genes   
   were and that they had regulatory sequences.  It made Mendelian genetics   
   pretty much totally understood.   
      
   As I mentioned we still do not understand why the infinite allele model   
   works so well in quantitative population genetics.  It is a model that   
   was needed to facilitate computation so that we could actually produce   
   some answers, but everyone understands that there are a finite number of   
   genes far fewer than anything close to infinite.  The human genome might   
   only have 15,000 coding genes, but as I also mentioned it looks like we   
   may not fully understand the biology, and the actual biology may make   
   the genome appear to generate a near infinite number of apparent   
   alleles.  We haven't worked out all of the biology.  We need to get a   
   better understanding of how dominance and gene interactions really   
   affect the population genetics.  Quantitative genetics currently do not   
   have adequate means for the analysis of dominance and epistasis (gene   
   interactions).  Nearly all the papers looking for the effects of   
   dominance and epistasis in populations under selection conclude that the   
   effects are minimal, but that is likely not true.   
      
   >   
   > I've mentioned here previously a partially completed computer program to   
   > simulate a population of ~10,000 "genomes", subject to sexual   
   > reproduction, with chromosomes, recombination and crossover, controlled   
   > randomisation, and mutations using various selection coefficient   
   > profiles, etc. I've been curious to attempt to explore fixation, viable   
   > selection coefficient distributions, genetic load and so on.   
      
   Probably all such programs that track alleles and assign selection   
   coefficients to alleles do not model how the genome actually evolves.   
      
   In reality selection coefficients and genetic load assignments have to   
   change as the allele frequencies in the populations change, but we   
   currently do not have a good way to do that, nor can we predict which   
   ones need to change with time and allele frequency.   
      
   The selection coefficient of a specific genotype can be dependent on the   
   allele frequency at another locus, so background genetics matter as well   
   as the environment.   
      
   Humans average a genetic load of 1.5 to 2.  These are lethal   
   equivalents, so if you were homozygous for your genetic load you would   
   be dead.  Drosophila studies indicate that over 50% of the genetic load   
   is due deleterious loci that are only 10% lethal or less when   
   homozygous.  10% lethal is just the percentage reduction from the   
   expected frequency of homozygotes in the population.  So in test crosses   
   where you expect 50% homozygotes you only find 45% homozygotes.  The   
   sporadic lethality of such homozygotes is likely due to environmental   
   influence, interaction with other sublethals, or deleterious   
   interactions with normally non lethal variants segregating in the   
   population.   
      
   How we calculate the lethal load and identify lethal loci is not that   
   accurate, mainly due to gene interactions messing with identification   
   and quantifying the lethal load.  There are multiple examples of a fully   
   lethal recessive trait associated with one loci that when crossed into   
   another genetic background is not lethal or incompletely lethal (only a   
   fraction of the homozygotes die).  Natural selection occurs in some   
   lines kept to carry recessive lethals so eventually the homozygotes do   
   not die.  Other loci in the genome were selected that counteracted the   
   lethality.   
      
   We also know that we have issues because of recombinant inbred lines.   
   You can take two to 6 highly inbred mouse lines that have all been   
   inbred long enough to be over 99% inbred (99% of the alleles are   
   identical by descent and homozygous).  These lines can have been   
   selected to be more reproductively successful than wild-type (more   
   litters and more pups per litter).  It could be claimed that the lethal   
   load in these lines was zero.  In the case of two inbred lines what they   
   do is cross them together and then backcross to one of the lines several   
   times so that they start mating full sibs that have different parts of   
   the donor genome and subsequent inbreeding produces lines where 12.5% of   
   the donor genome is fixed (homozygous) dispersed around the genome.   
   Each recombinant inbred line has a different 12.5% of the donor genome   
   so you can use around 20 recombinant inbred lines to genetically map   
   variants from the donor genome.  The problem is that many of these   
   recombinant inbred lines start to fail to reproduce enough progeny to   
   maintain the recombinant inbred lines.  It turns out that parts of the   
   donor genome has a lethal load when combined with the genetic background   
   of the other highly inbred line.  They lose the lines if they continue   
   to inbreed them, so they start maintaining the lines as inbred as they   
   can make them, but they remain heterozygous for parts of the donor genome.   
      
   This just means that all the simplistic models of assigning genetic load   
   and selection coefficients to genotypes are inadequate to model what   
   actually happens.   
      
   >   
   > Inconclusive so far, but the exercise has been an impetus to try and   
   > understand some of the principles involved. I plan to get back to it,   
   > but further study of pop gen first would help verify my assumptions and   
   > modelling.   
   >   
   > My initial approach was to start with a supposed selection coefficient   
   > distribution. The data I could find suggested some lethal (-1), some   
   > deleterious (-1 < x < near-neutral), many neutral or near-neutral (zero   
   > or just under), and a small number beneficial (just above zero). Using   
   > this, determine if the population grows or goes extinct through genetic   
   > load. However, I found it difficult to find definitive data, and so   
   > instead flipped the approach to reverse-engineer a "break-even"   
   > selection coefficient distribution.   
   >   
   > One question that presents itself is how to model overall relative   
   > fitness of an individual carrying multiple mutations. The simple   
   > solution is to just add them together. Of course, in nature complex and   
   > dynamic non-linear effects apply, which are beyond a simple simulation.   
   > However, it seems to me that a well-constructed simulation could give a   
   > reasonably indicative picture.   
   Doing something like this right takes a lot of advanced modeling and   
   dealing with biology that we haven't yet completely worked out.  We know   
   that things like gene interaction and dominance need to be in the   
   models, but we don't know how much, nor can we predict when these things   
   are factors.   
      
   I do think that evolutionary models need to deal with the genetic load   
   and inbreeding.  I believe that the deleterious genetic load in the   
   population is very important in maintaining genetic variation and   
   selection progress in a population.  It may be a major factor in why   
   populations fail, and why others take over.  Why is the spotted owl   
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)