Beacham: The use of DNA variation for stock identification of Oncorhynchus keta 



615 



functions were developed with all fish 

 sampled except the one to be classified. The 

 accuracy of classification of individual fish 

 was summarized both on a population and 

 on a regional basis. 



Estimation of stock 

 composition 



Baseline data 



In this study, counts were available from 

 15 alleles at the Ssa-A33 locus, 18 alleles 

 at the Ssa -A34 locus, and from 31 band 

 counts at the Ssal locus for each fish. The 

 maximum-likelihood model (Fournier et al., 

 1984) used in the estimation of stock com- 

 position requires that the characters used 

 in the analysis be independent. When al- 

 lele counts and band counts were used to 

 characterize the chum salmon stocks, a 

 principal-components analysis on the cor- 

 relation matrix of the 64 variables in the 

 study was conducted in order to obtain 

 uncorrelated variables or principal compo- 

 nents, as was described by Beacham et al. 

 ( 1996). Summarized briefly, the PRINCOMP 

 procedure in SAS was used in the princi- 

 pal-components analysis. The observed data 

 (number of bands per bin) were standard- 

 ized by subtracting the overall correspond- 

 ing mean and by dividing by the correspond- 

 ing bin standard deviation over all popula- 

 tions for the original variables. The origi- 

 nal variables were then represented as fac- 

 tor scores for uncorrelated variables or prin- 

 cipal components. The first 57 principal 

 components accounted for 100% of the to- 

 tal observed variation. The input to the 

 maximum-likelihood model required dis- 

 crete frequency counts for each variable, 

 and as outlined by Beacham et al. (1995), 

 the continuous distributions of the princi- 

 pal component scores were represented as 

 12-bin histograms. Limits of the bins were 

 as follows: <-2.50 (1), -2.49 to -1.50 (2), - 

 1.49 to -1.00 (3), -0.99 to -0.60 (4), -0.59 to 

 -0.30 (5), -0.29 to 0.00 (6), 0.00 to 0.29 (7), 

 0.30 to 0.59 (8), 0.60 to 0.99 (9), 1.00 to 1.49 

 (10), 1.50 to 2.49 (11), and >2.50 (12). Each 

 stock was then characterized as an input 

 matrix of 57 rows (principal components) 

 and 12 columns (counts of principal compo- 

 nent scores in each of the 12 bins). 



