SPAM

Version 3.7 Addendum II to Users Guide for Version 3.2

STATISTICS PROGRAM FOR ANALYZING MIXTURES

Special Publication No. 15 Alaska Department of Fish and Game Division of Commercial Fisheries Gene Conservation Laboratory 333 Raspberry Road Anchorage, Alaska 99518

25 June 2003

(this page intentionally blank)

iv

SPAM
(Statistics Program for Analyzing Mixtures) Version 3.7 Addendum II to Users Guide for Version 3.2

Addendum II to Special Publication No. 15 Alaska Department of Fish and Game Division of Commercial Fisheries Gene Conservation Laboratory 333 Raspberry Road Anchorage, Alaska 99518 25 June 2003

Citing the Software: SPAM 3.2: Debevec, E. M., R. B. Gates, M. Masuda, J. Pella, J. Reynolds, L. W. Seeb. 2000. SPAM (Version 3.2): Statistics Program for Analyzing Mixtures. Journal of Heredity 91 (6):509-510. SPAM 3.5: Alaska Department of Fish and Game. 2001. SPAM Version 3.5: Statistics Program for Analyzing Mixtures. Alaska Department of Fish and Game, Commercial Fisheries Division, Gene Conservation Lab. Available for download from http://www.cf.adfg.state.ak.us/geninfo/research/genetics/Software/SpamPage.htm. Citing the Manual: SPAM 3.2: Alaska Department of Fish and Game. 2000. SPAM Version 3.2: Users Guide. Special Publication 15, Alaska Department of Fish and Game, Commercial Fisheries Division, Gene Conservation Lab, 333 Raspberry Road, Anchorage, Alaska, 99518. 61 pages. Available for download from http://www.cf.adfg.state.ak.us/geninfo/research/genetics/Software/SpamPage.htm. SPAM 3.5: Reynolds, J. H. 2001. SPAM Version 3.5: Users Guide Addendum. Addendum to Special Publication 15, Alaska Department of Fish and Game, Commercial Fisheries Division, Gene Conservation Lab, 333 Raspberry Road, Anchorage, Alaska, 99518. 63 pages. Available for download from http://www.cf.adfg.state.ak.us/geninfo/research/genetics/Software/SpamPage.htm.

Product names used in this publication are included for scientific completeness but do not constitute product endorsement. Microsoft and Windows are registered trademarks of Microsoft Corporation.

ii

Table of Contents

Overview of Enhancements ............................................................................................ 1 New Control File Options............................................................................................... 3 Estimation/Simulation................................................................................................. 3 Options........................................................................................................................ 3 Rannala-Mountain................................................................................................... 4 Pella-Masuda........................................................................................................... 4 Parameters................................................................................................................... 5 Characters ................................................................................................................... 6 Populations.................................................................................................................. 6 Regions ....................................................................................................................... 6 Files............................................................................................................................. 6 Run.............................................................................................................................. 7 Input Files ....................................................................................................................... 8 Output Files..................................................................................................................... 9 Estimation (*.est) ..................................................................................................... 9 Simulation (*.sim) ..................................................................................................... 9 Bootstrap (*.bot) ..................................................................................................... 10 Baseline (*.bsl)....................................................................................................... 10 Conditional Genotype Probabilities (*.gen) ............................................................ 10 Correspondence............................................................................................................. 11 Limited Warranty and Disclaimer ................................................................................ 12 Literature Cited ............................................................................................................. 13 Appendix....................................................................................................................... 14 Example Control File................................................................................................ 14

iii

(this page intentionally blank)

iv

Overview of Enhancements

This addendum describes the new features of SPAM version 3.7. For a full description of the program, including input and output files, please see the SPAM version 3.2 Users Guide and the SPAM version 3.5 Addendum to the Users Guide, both available online1. The following descriptions assume a familiarity with material in the Users Guide and Addendum, hereafter referred to as UG:3.2. The enhancements in SPAM 3.7 accommodate the estimation of baseline allele frequencies for loci with many low-frequency alleles. Under the conditional maximum likelihood (CML) scheme, a stock that has a sampled frequency of zero for an allele is assumed to be an impossible source for a mixture individual with that allele. This assumption may be unrealistic and cause bias and/or imprecision in stock-composition estimates. SPAM 3.7 offers a solution to this problem by allowing Bayesian modeling of baseline allele frequency distributions within the maximum likelihood scheme. That is, maximum likelihood is still used to estimate stock composition. SPAM 3.7 offers two Bayesian models of baseline allele frequency distributions: 1) Rannala-Mountain (Rannala and Mountain 1997) and 2) Pella-Masuda baseline posteriors (Pella and Masuda 2001). Both baseline posteriors are Dirichlet distributions. Rannala and Mountain (1997) use an equal-probability prior distribution (their Eq. 1) for the alleles at a locus with mean frequency equal to one over the number of distinct alleles. That is, all alleles at a locus are assumed to be equally abundant for all stocks before the baseline samples become available. The mean of the Rannala and Mountain baseline posterior is an unweighted average of the observed allele relative frequencies and the prior mean (Eq. 6 of Rannala and Mountain 1997). Pella and Masuda (2001) use a pseudo-Bayes method to determine the baseline posterior distribution for the alleles at a locus. The baseline center or unweighted arithmetic mean of the allele frequencies among stocks at a locus is used as the mean of the prior distribution. The mean of the Pella-Masuda baseline posterior is a weighted average of the observed allele relative frequencies and the baseline center with weights determined by an objective risk criterion (Eq. 4 of Pella and Masuda 2001). The user may still perform with SPAM 3.7 a traditional CML analysis, using the maximum likelihood estimates of baseline allele frequencies (i.e. no Bayesian modeling of baseline allele frequency distributions) if baseline sampling of zeros for alleles is not of concern. Assuming the Rannala-Mountain (or Pella-Masuda) model of baseline allele frequency distributions, the user may perform estimation (with or without bootstrapping) or simulation (with baseline resampling). If the analysis involves no baseline resampling (for bootstrapping or simulation), then baseline allele frequencies are estimated by the mean of the Dirichlet posterior distributions. If the analysis involves baseline resampling
1

http://www.cf.adfg.state.ak.us/geninfo/research/genetics/Software/SpamPage.htm

(for bootstrapping or simulation), then allele frequencies are repeatedly drawn from the Dirichlet posterior distributions. If simulations are done, mixture genotypes are generated by sampling alleles with fixed relative frequencies equal to the posterior mean.

New Control File Options

For full documentation regarding required input files, including the general options available in the control file, see the SPAM version 3.2 Users Guide and the SPAM version 3.5 Addendum to the Users Guide. * Estimation/Simulation No new options. * Options See Appendix for example control file. * options selected for optimization use IRLS algorithm in optimal search print mixture file print baseline relative frequencies print conditional genotype probabilities print conditional population probabilities print bootstrap estimates print iterations print likelihoods of simulations/resamples compute likelihood confidence intervals compute infinitesimal jackknife std. dev. compute studentized conf. intervals compute likelihood at external estimate resample mixture frequencies resample baseline use rannala-mountain model use pella-masuda model : : : : : : : : : : : : : : : : t t t t t t t t t t t t t t t f

This section is used to select performance and output options. The keywords for the various options are listed here. Keyword(s) IRLS PRIN BASE MIXT GENO POPU or STOC F F F F Default F Description Use IRLS algorithm in optimal search Print baseline relative frequencies mixture file conditional genotype probabilities conditional population (stock) probabilities 3

BOOT ITER LIKE COMP CONF JACK STUD EXTE

F F F

estimates from each bootstrap resample MLE search iterations likelihood for each simulation or resample Compute likelihood confidence intervals infinitesimal jackknife standard deviations studentized bootstrap conf. intervals use externally provided mle in likelihood ratio Resample mixture frequencies baseline Use Rannala-Mountain model of baseline allele frequency distributions Use Pella-Masuda model of baseline allele frequency distributions

F F F F

RESA MIXT BASE RANN F F F

PELL

RANNALA-MOUNTAIN The keyword RANNALA is used for modeling baseline allele frequency distributions with the Rannala-Mountain baseline posterior (Rannala and Mountain 1997). If the analysis involves no baseline resampling, then baseline allele frequencies are estimated by the mean of the Dirichlet posterior distributions. If the analysis involves baseline resampling, then allele frequencies are repeatedly drawn from the Dirichlet posterior distributions. PELLA-MASUDA The keyword PELLA is used for modeling baseline allele frequency distributions with the Pella-Masuda baseline posterior (Pella and Masuda 2001). If the analysis involves no baseline resampling, then baseline allele frequencies are estimated by the mean of the Dirichlet posterior distributions. If the analysis involves baseline resampling, then allele frequencies are repeatedly drawn from the Dirichlet posterior distributions. If both keywords RANNALA and PELLA are set to true, then the program will default to the Rannala-Mountain model.

* Parameters See Appendix for example control file. The control parameters specify the number of populations and characters in the analysis, upper limit parameters, tolerances to control the optimization search, and features to synchronize and partition simulated mixture samples across runs of SPAM 3.5 ( for use in Monte Carlo simulation of likelihood ratios for testing mixture equality). * control parameters number of populations in analysis number of characters in analysis maximum number of genotypes maximum number of classes maximum # of iterations maximum number of missing loci estimate tolerance likelihood tolerance genotype tolerance algorithm switch tolerance GPA number of resamplings simulation sample size number of null observations after confidence intervals random seed (negative) second random seed (positive) The keywords for the commands are listed here: Keyword(s) Default NUMB POPU or STOC CHAR RESA 100 BEFO 0 AFTE 0 Description Number of populations (stocks) in the analysis characters in the analysis bootstrap resamplings observations to simulate, but not use, before simulating mixture sample of interest observations to simulate, but not use, after simulating mixture sample of interest Maximum number of genotypes classes iterations missing (unscored) loci in mixture Tolerances for estimates : : : : : : : : : : : : : : : : : 14 9 200 20 300 4 .1E-3 1.0E-10 1.0E-6 0.01 90 100 100 0 90 -718805 99733654

MAXI GENO CLAS ITER MISS TOLE ESTI 10-4 100 1 100 0

LIKE or FUNC GENO ALGO or SWIT GUAR, PERC, or GPA CONF SIZE SEED SECO

10-10 10-10 10-2 90

likelihood (function) genotype probability algorithm switch (CG to IRLS) Guaranteed percent achievement of the maximal likelihood (GPA) Confidence interval size (percent) Simulation sample size Random number generator seed Random number generator seed

90 100 From CPU clock From CPU clock

SECOND The keyword SECOND allows the user to specify a seed for a second random number generator for reproducible results. The seed takes a positive value between 0 and 2,147,483,648. If a negative seed is given, it will be changed to positive. Therefore, the seed 12,345 will give the same sequence of numbers as 12,345. The seed will be printed in the bootstrap and/or simulation output files. If a random seed is not declared in the control file, SPAM will generate a seed based on the current CPU time. The seed will be printed in the bootstrap and/or simulation output files for reference. * Characters No changes. See UG:3.2 for full description. * Populations No changes. See UG:3.2 for full description. * Regions No changes. See UG:3.2 for full description. * Files No changes. See UG:3.2 for full description. 6

* Run No changes. See UG:3.2 for full description.

Input Files

No changes. See UG:3.2 for full description.

Output Files

All results from a SPAM analysis are printed to a collection of ASCII text files that can be viewed through the SPAM environment or separately with any text editor. The set of files created depends on the analysis requested in the control file. All files, except the resampled estimate files, are formatted for convenient viewing and printing. Every SPAM analysis will produce a log file (*.log) and either an estimation (*.est) or a simulation file (*.sim), depending on the type of analysis run. Only changes in content or new files are discussed below. See UG:3.2 for a full description of the other output files created by SPAM 3.7. Estimation (*.est) See UG:3.2 for a full description. The file will indicate if either the Rannala-Mountain or Pella-Masuda model of baseline allele frequency distributions is used. If the Pella-Masuda model is selected, then values for the baseline posterior parameters are determined from the pseudo-Bayes method described in Pella and Masuda (2001). The posterior mean of allele or type relative frequencies is computed as a weighted average of the observed and prior mean relative frequencies. Weights for prior means are output to the *.est file for each character and stock. Weights for observed means are simply one minus weights for prior means. Large weights for prior means may indicate that there is little variation in the character among stocks. The characters value in discriminating stocks should be examined. Simulation (*.sim) See UG:3.2 for a full description. The file will indicate if either the Rannala-Mountain or Pella-Masuda model of baseline allele frequency distributions is used in the resampling of the baseline. The value of the second random seed (positive) is reported here. If the Pella-Masuda model is selected, then values for the baseline posterior parameters are determined from the pseudo-Bayes method described in Pella and Masuda (2001). The posterior mean of allele or type relative frequencies is computed as a weighted average of the observed and prior mean relative frequencies. The posterior mean is the underlying fixed allele or type relative frequencies of the simulated populations. Weights 9

for prior means are output to the *.sim file for each character and stock. Weights for observed means are simply one minus weights for prior means. Large weights for prior means may indicate that there is little variation in the character among stocks. The characters value in discriminating stocks should be examined. Bootstrap (*.bot) See UG:3.2 for a full description. The file will indicate if either the Rannala-Mountain or Pella-Masuda model of baseline allele frequency distributions is used in the resampling of the baseline. The value of the second random seed (positive) is reported here. Baseline (*.bsl) See UG:3.2 for a full description. If either the Rannala-Mountain or Pella-Masuda model is selected, the means of the Dirichlet posterior distributions of allele frequencies are printed below the observed baseline allele frequencies. Conditional Genotype Probabilities (*.gen) See UG:3.2 for a full description. If either the Rannala-Mountain or Pella-Masuda model is selected, conditional probabilities of types found in the mixture are computed from baseline allele frequencies estimated by the mean of the Dirichlet posterior distributions.

10

Correspondence

We welcome correspondence regarding SPAM. If you would like to be included on the mailing list and receive notifications of updates, please contact us at the address below. Please report any bugs as soon as possible so we can assess the problem and make any necessary corrections to the program. Bill Templin [email protected] Lisa Seeb [email protected] Alaska Department of Fish and Game Division of Commercial Fisheries Gene Conservation Laboratory 333 Raspberry Road Anchorage, Alaska 99518 USA

11

Limited Warranty and Disclaimer

This software and accompanying written materials (including instructions for use) are provided as is without warranty of any kind. Further, Alaska Department of Fish and Game (ADF&G) does not warrant, guarantee, or make any representations regarding the use, or the results of use, of the software or written materials in terms of correctness, accuracy, reliability, currentness, or otherwise. The entire risk as to the results and performance of the software is assumed by you. If the software or written materials are defective, you, and not ADF&G or its employees, assume the entire cost of all necessary servicing, repair, or correction. The above is the only warranty of any kind, either express or implied, including but not limited to the implied warranty of fitness for a particular purpose, that is made by ADF&G. No oral or written information or advice given by ADF&G or its employees shall create a warranty or in any way increase the scope of this warranty and you may not rely on any such information or advice. Neither ADF&G nor anyone else who has been involved in the creation, production or delivery of this product shall be liable for any direct, indirect, consequential or incidental damages (including damages for loss of business profits, business interruption, loss of business information, and the like) arising out of the use or inability to use such product even if ADF&G has been advised of the possibility of such damages. Use of this product for any period of time constitutes your acceptance of this agreement and subjects you to its contents.

12

Literature Cited

Pella, J. and Masuda, M. 2001. Bayesian methods for analysis of stock mixtures from genetic characters. Fishery Bulletin. 99:151-167. Rannala, B. and Mountain, J. L. 1997. Detecting immigration by using multilocus genotypes. Proceedings of the National Academy of Sciences of the United States of America. 94:9197-9201.

13

Appendix

Example Control File for Modeling Baseline Allele Frequency Distributions. * estimation: Columbia Example * options selected for optimization use IRLS algorithm in optimal search print mixture file print baseline relative frequencies print conditional genotype probabilities print conditional population probabilities print bootstrap estimates print iterations compute likelihood confidence intervals compute infinitesimal jackknife std. dev. compute studentized bootstrap intervals resample mixture frequencies resample baseline use rannala-mountain model use pella-masuda model * control parameters number of populations in analysis number of characters in analysis maximum number of genotypes maximum number of classes maximum # of iterations maximum number of missing loci estimate tolerance likelihood tolerance genotype tolerance algorithm switch tolerance GPA number of resamplings confidence intervals random seed second random seed (positive) ... : : : : : : : : : : : : : : : : : : : : : : : : : : : : : f t t t t t t t t t f f t f

14 9 200 20 300 4 .1E-3 1.0e-10 1.0e-6 0.01 90 25 90 -718805 85728283

14

The Alaska Department of Fish and Game administers all programs and activities free from discrimination based on race, color, national origin, age, sex, religion, marital status, pregnancy, parenthood, or disability. The department administers all programs and activities in compliance with Title VI of the Civil Rights Act of 1964, Section 504 of the Rehabilitation Act of 1973, Title II of the Americans with Disabilities Act of 1990, the Age Discrimination Act of 1975, and Title IX of the Education Amendments of 1972. If you believe you have been discriminated against in any program, activity, or facility, or if you desire further information please write to ADF&G, P.O. Box 25526, Juneau, AK 99802-5526; U.S. Fish and Wildlife Service, 4040 N. Fairfield Drive, Suite 300, Arlington, VA 22203 or O.E.O., U.S. Department of the Interior, Washington DC 20240. For information on alternative formats for this and other department publications, please contact the department ADA Coordinator at (voice) 907-465-4120, (TDD) 907-465-3646, or (FAX) 907-465-2440.