xia@cc.helsinki.fi (10/21/90)
A Quick and Dirty Method for Measuring Gene Flow ================================================ Let P(A) represent frequency of allele A of a neutral polymorphic locus. For n populations, let P1t(A) be the frequency of A in population 1 at generation t, P2t(A) be the frequency of A in population 2 at generation t,......, and Pnt(A) be the frequency of A in population n at generation t. Let P1_t+1(A), P2_t+1(A),......, Pn_t+1(A) be the corresponding allelic frequency at generation t+1. Now we have the following table. ------------------------------------ PopulationPt(A) Pt+1(A) 1 P1t(A) P1_t+1(A) 2 P2t(A) P2_t+1(A) 3 P3t(A) P3_t+1(A) . . . . . . n Pnt(A) Pn_t+1(A) ------------------------------------ The relationship between Pt+1(A) and Pt(A) is Pt+1(A) = a + b*Pt(A) + E. When there is neither gene flow nor random factors such as drift, we have a linear relationship with intercept=0, slope=1 and E=0. When there is only random drift, we have a linear relationship with a < 0, b > 1 and E<>0 (not equal to 0). This is because the probability of an allelic frequency, say 0.8, drifting to 1 is always greater than the probability of an allelic frequency of 1 drifting to 0.8 (The latter probability is 0). Similarly, the probability of an allelic frequency of 0.2 drifting to 0 is always greater than the probability of an extinct allele drifting back to 0.2, the latter probability again being 0. In this special case, b-1 (a positive value) is a measure of random drift. When there is only gene flow, we have a linear relationship with a > 0, b < 1 and E<>0 (not equal to 0). This is because the probability of an allelic frequency of 1 becoming smaller than 1 is always greater than that of becoming larger than 1, if only gene flow is involved. Similarly, the probability of an allelic frequency of 0 becoming larger than 0 is always greater that that of becoming smaller than 0. In this special case, b-1 (a negative value) is a measure of gene flow. When there are both gene flow and random drift, then we need to calculate b not only for allele A of one particular locus, but also for many other loci. Although allelic frequency of most loci may be under control of both random drift and gene flow, some loci may be affected only by random drift and some only by gene flow. Thus, if we obtain b for each of many loci, then the largest b value minus 1 is a measure of random drift and the smallest b value minus 1 is a measure of gene flow. This method apparently is not affect by historical factors. Please note that the validity and utility of the method has not been checked. I propose it just to be criticized. Xuhua