Linkage Newsletter

Jurg Ott jo7 at namaste.cc.columbia.edu
Tue Aug 2 13:47:11 EST 1994


                       Linkage Newsletter

                  Vol. 8   No. 2   August 1994

Published by Jurg Ott, Columbia University, New York

Editorial Assistant:  Katherine Montague
     Fax: 212-568-2750
     Tel. 212-960-2507
     e-mail:  jurg.ott at columbia.edu

Postal address:
     Columbia University, Unit 58
     722 West 168th Street, New York, NY 10032


E-mail address

     Our old bitnet address (ott at nyspi) is no longer valid and
will soon reject mail sent to it.  If you want to send us e-mail,
please use only the address given above.  Also, for those who use
e-mail but receive this newsletter by postal mail, please let us
know -- we would prefer sending the newsletter by e-mail only.


LINKAGE COURSES

     The next two linkage courses are Advanced Courses.  They
will take place as follows:
     October 3-7, 1994, at the University of Zurich, Irchel
Campus Computer Center (Switzerland;  maximum number of par-
ticipants is 14), and
     January 9-13, 1995, at Columbia University, New York (maxi-
mum of 20 participants).
     To obtain information on these courses, please write to
Katherine Montague, course coordinator, by e-mail or fax.  These
courses are for researchers with experience in using the LINKAGE
programs, or who have an otherwise excellent understanding of
genetic linkage analysis.  Also, course participants must be
familiar with PCs.  Topics:  Working with age at disease onset; 
models for genetic heterogeneity;  genetic linkage and allelic
association;  complex traits;  etc.  As usual, the main focus
will be on practical exercises and linkage analyses carried out
by participants on IBM PC's using the LINKAGE and other programs. 
Each session will begin with a theoretical introduction on the
material to be worked on.  We will use our new book instead of
mimeographed handouts (Terwilliger and Ott, "Handbook for Human
Genetic Linkage," Johns Hopkins University Press, 1994).  Next
October, the book will be handed out at the course but for later
courses (eg. January 1995), participants are expected to buy the
book and bring it to the course.

     The next introductory courses will be held in the spring or
early summer of 1995, one each at Columbia University New York
and at the University of Zurich (dates not yet set).


SOFTWARE NEWS

PC version 5.2 of LINKAGE

     Because of as yet unresolved problems with version 5.2 of
some LINKAGE programs (PC, Turbo Pascal), for general pedigrees
we are now distributing version 5.1 while for the three-generat-
ion pedigrees version 5.2 is used.

FastSLINK

     As mentioned in the last newsletter, Drs. Daniel Weeks and
Alejandro Schaffer have improved SLINK so that it runs consider-
ably faster than the original version.  This C-version is now on
Dr. Week's ftp server, watson.hgen.pitt.edu.  A brief description
of the files available on that server follows:

------------------------------------------------------------------------
readme           :  Readme file for this directory
newapm.readme    :  Readme about the new version of the APM programs
newdosapm.tar.Z  :  DOS version of the new APM programs
newdosapm.zip    :  DOS version of the new APM programs
newdosapm.readme :  Readme about the DOS version of the new APM programs
newdosapm.tar.Z  :  DOS version of the new APM programs
slink.tar.Z      :  SLINK simulation program (original version) Weeks DE,
                    Ott J, Lathrop GM (1990) Am J Hum Genet 47:A204
fastslink.tar.Z  :  Fast version of SLINK.  SLINK as modified by Alejandro
                    Schaffer and Daniel Weeks to use the algorithms in
                    FASTLINK v. 1;  cf. Cottingham Jr RW, Idury RM, Schaf-
                    fer AA, Am J Hum Gen 53(1993), 252-263
cintmax.tar.Z    :  Linkage analysis under mapping functions
simapm.tar.Z     :  SLINK-based simulation program for APM method (Unsta-
                    ble Unix-hack -- use at own risk)
------------------------------------------------------------------------

Borland Pascal version 7.0

     We are gradually converting all DOS Pascal programs to
Borland Pascal 7.  The main advantage is that programs may be
compiled to run in protected mode so that they can make use of
extended memory.  Thus, for many users a heap space overflow will
no longer be a problem.  Programs may still be compiled to run in
real mode (as in earlier versions of Turbo Pascal) in which case
execution speed will be about 10% higher than when they run in
protected mode.  In either case, however, the usual restrictions
of Turbo/Borland apply, in particular, the data segment is still
limited to 64KB.  Program version without these limitations are
available for OS/2 (NDP Pascal).

Bugs in CFACTOR program

     In the three-generation programs (CLINKAGE), pedigree input
files are preprocessed by the CFACTOR program such that larger
numbers (20 or more on the PC) loci may be analyzed jointly.  Dr.
Weeks has reported two bugs that will result in somewhat errone-
ous likelihoods.  The bugs and their suggested corrections are
given below.

== Bug #1 ==

     In each of the two procedures, getpaphase and getmaphase,
the following two lines occur:

     gpahet:=(pa^.gene1[i]<>pa^.gene2[i]) and gpaknown;
     gmahet:=(pa^.gene1[i]<>pa^.gene2[i]) and gpaknown; <=BUG 

The second line must be corrected in that all occurrences of 'pa'
in the buggy lines should be changed to 'ma', since everything
should refer to the grandmother, rather than to the grandfather. 
So, the second line, in each of the two procedures, should be

     gmahet:=(ma^.gene1[i]<>ma^.gene2[i]) and gmaknown; <=CORR

== Bug #2 ==

     The procedure getdoublef factorizes (Lathrop, Lalouel, and
White 1986) pedigrees in specific situations.  Evidently, it does
this also in cases where factorization should not be applied. 
The complete discussion of this bug is quite elaborate and is not
reproduced here.  Dr. Weeks recommends as a solution that the two
lines indicated below should be deleted, that is, the curly
brackets {} should be inserted in the lines identified by DELETE:

     procedure getdoublef;
     {Doublef contains information on factorization
      of completely informative loci}
     var
       i,j : integer;
       paside1,paside2,maside1,maside2 : boolean;
     begin
       FOR i:=1 TO nlocus DO doublef[i]:=false;
       FOR i:=2 TO nlocus-1 DO
         IF pahet[i] and mahet[i] and not intercross[i] THEN
           IF paphase[i] and maphase[i] THEN doublef[i]:=true
           ELSE
             {Factorize if phase unknown for surrounding loci
              on at least one side for each parent}
             begin
               paside1:=false;
               paside2:=false;
               maside1:=false;
               maside2:=false;
           {   IF not paphase[i] THEN  <= DELETE }
                 begin
                   FOR j:=1 TO i-1 DO
                     IF paphase[j] THEN paside1:=true;
                   FOR j:=i+1 TO nlocus  DO
                     IF paphase[j] THEN paside2:=true;
                 END;
           {   IF not maphase[i] THEN  <= DELETE }
                 begin
                   FOR j:=1 TO i-1 DO
                     IF maphase[j] THEN maside1:=true;
                   FOR j:=i+1 TO nlocus  DO
                     IF maphase[j] THEN maside2:=true;
                 END;
               IF not ((paside1 and paside2)
                   or (maside1 and maside2)) THEN doublef[i]:=true;
             END;
     END;


Bug in LINKMAP program

     The report presented below has been submitted by Dr. Alejan-
dro Schaffer.  Over a year ago, an analogous report was sent to
me by Dr. Weeks but at the time I considered it to be more a nui-
sance than a bug;  at least it is an inelegant aspect of the
LINKMAP program.  Dr. Schaffer's report is as follows:

     There is a bug in LINKMAP in LINKAGE 5.1 and LINKAGE 5.2 in
which components of the theta vector which are supposed to be 0.0
are slightly positive.  This occurs when that component is de-
creased from a larger positive number (e.g., 0.5) down to 0.0. 
What is particularly nasty about this bug is that whether it
occurs or not depends on the number of steps of decrease as shown
in the output transcript below.  Note the non-infinite answer in
the last run for the theta vector 0.112 0.210 0.000 in contrast
to the infinite answer for what appears to be the SAME theta
vector in the first two runs.  The only difference between the
runs is how many steps of  decrease are used to reduce from 0.210
to 0.000.  As my example shows, a bad consequence of this bug is
that when the pedigree data implies that there must be a recombi-
nation and having the theta component be 0.0 should give a
-infinity log likelihood, you instead can get a semi-plausible
negative number.  Furthermore, having the actual value be nonzero
(when it should be zero) drastically slows down the computation
for that theta.  I have repaired the bug in LINKMAP and MLINK for
what will be FASTLINK 2.2.  I am not ready to release FASTLINK
2.2 because I want to take care of some other things that users
requested.
     NB: The transcript below was prepared with FASTLINK 2.1
(which still has the bug).  I checked that LINKAGE 5.1 and 5.2
give the same buggy result on a Sparcstation 10.
     Caution: The occurrence of this bug may be architecture-
dependent as testing for 0.0 is well known to be a hard problem.

     The program output referred to above is given in the Appen-
dix.

Bug in LOOPS program

     Dr. Weeks reported a bug in the LOOPS program (OS/2 version
only).  Close to line 500, the following two lines occur:

     readln(loopfile);
     writeln(loopfile);

Correction:    The second line should be changed to
               writeln(outfile).

     Note (J.O.):  This error evidently occurred in the adap-
tation from Turbo Pascal to NDP Pascal as it is not present in
the DOS version.  I am grateful to Dr. Weeks for pointing it out
to me.

No bug in Vax version of LINKAGE - and a word of caution

     Over a year ago a 'bug' was reported here describing an
apparent difference between the Vax and DOS versions of LINKAGE. 
The Vax version seemed to incorrectly produce a lod score of zero
for a single doubly homozygous offspring of a phase-unknown
double intercross mating in the presence of allelic association. 
The reason for the discrepancy between the two versions has now
been traced to the program constant FITMODEL - it was set to
false in the Vax version and to true in the DOS version.  In the
program description, this constant is said to have an effect in
the ILINK program when more than just recombination fractions are
to be estimated.  It evidently has other effects as well.  So, it
is prudent to set FITMODEL equal to true at all times except in
special circumstances.

Support through grant HG00008 from the National Center for
Human Genome Research is gratefully acknowledged.


APPENDIX

***************************************************************************
Run   1 - LINKMAP : p12 p14 p1 p17
***************************************************************************
Program UNKNOWN version 5.1 (1-Feb-1991)
The following maximum values are in effect:
      10 loci
      55 single locus genotypes
      10 alleles at a single locus
     500 individuals in one pedigree
       3 marriage(s) for one male
       3 quantitative factor(s) at a single locus
      20 liability classes
      10 binary codes at a single locus
YOU ARE USING LINKAGE (V5.1 (1-Feb-1991)) WITH  4-POINT AUTOSOMAL DATA
Program LINKMAP version  5.10 (1-Feb-1991)
FASTLINK version  2.10 (21-Mar-1994)
The program constants are set to the following maxima:
     8 loci in mapping problem (maxlocus)
    12 alleles at a single locus (maxall)
   157 recombination probabilities (maxneed)
 50000 maximum of censoring array (maxcensor)
   180 haplotypes = n1 x n2 x ... where ni = current # alleles at locus i
 16290 joint genotypes for a female
 16290 joint genotypes for a male
  1000 individuals in all pedigrees combined (maxind)
   150 pedigrees (maxped)
    10 binary codes at a single locus (maxfact)
     3 quantitative factor(s) at a single locus
    20 liability classes
     3 quantitative factor(s) at a single locus
    20 liability classes
    10 binary codes at a single locus
    2.00 base scaling factor for likelihood (scale)
    2.00 scale multiplier for each locus (scalemult)
 0.00000 frequency for elimination of heterozygotes (minfreq)
YOU ARE USING LINKAGE (V5.1 (1-Feb-1991)) WITH  4-POINT
YOU ARE USING FASTLINK (V2.1 (21-Mar-1994)) AUTOSOMAL DATA
Number of alleles at locus 1 is 6
Number of alleles at locus 2 is 5
Number of alleles at locus 3 is 2
Number of alleles at locus 4 is 3
-----------------------------------
-----------------------------------
THETAS  0.112 0.000 0.210
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
        1  -203.121527   -88.214370
-----------------------------------
TOTALS     -203.121527   -88.214370
-2 LN(LIKE) =  4.06243E+02
Maxcensor can be reduced to       -32767
-----------------------------------
-----------------------------------
THETAS  0.112 0.210 0.000
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
        1 -100000000000000000000.000000 -43429355638650388480.000000
-----------------------------------
TOTALS    -100000000000000000000.000000 -43429355638650388480.000000
-2 LN(LIKE) =  2.00000E+20
Maxcensor can be reduced to       -32767
-2 LN(LIKE) =  2.00000E+20
Maxcensor can be reduced to       -32767
***************************************************************************
Run   1 - LINKMAP : p12 p14 p1 p17
***************************************************************************
Program UNKNOWN version 5.1 (1-Feb-1991)
The following maximum values are in effect:
      10 loci
      55 single locus genotypes
      10 alleles at a single locus
     500 individuals in one pedigree
       3 marriage(s) for one male
       3 quantitative factor(s) at a single locus
      20 liability classes
      10 binary codes at a single locus
      20 liability classes
      10 binary codes at a single locus
YOU ARE USING LINKAGE (V5.1 (1-Feb-1991)) WITH  4-POINT AUTOSOMAL DATA
Program LINKMAP version  5.10 (1-Feb-1991)
FASTLINK version  2.10 (21-Mar-1994)
The program constants are set to the following maxima:
     8 loci in mapping problem (maxlocus)
    12 alleles at a single locus (maxall)
   157 recombination probabilities (maxneed)
 50000 maximum of censoring array (maxcensor)
   180 haplotypes = n1 x n2 x ... where ni = current # alleles at locus i
 16290 joint genotypes for a female
 16290 joint genotypes for a male
 16290 joint genotypes for a female
 16290 joint genotypes for a male
  1000 individuals in all pedigrees combined (maxind)
   150 pedigrees (maxped)
    10 binary codes at a single locus (maxfact)
     3 quantitative factor(s) at a single locus
    20 liability classes
    10 binary codes at a single locus
    2.00 base scaling factor for likelihood (scale)
    2.00 scale multiplier for each locus (scalemult)
 0.00000 frequency for elimination of heterozygotes (minfreq)
YOU ARE USING LINKAGE (V5.1 (1-Feb-1991)) WITH  4-POINT
YOU ARE USING FASTLINK (V2.1 (21-Mar-1994)) AUTOSOMAL DATA
Number of alleles at locus 1 is 6
Number of alleles at locus 2 is 5
Number of alleles at locus 3 is 2
Number of alleles at locus 4 is 3
Number of alleles at locus 3 is 2
Number of alleles at locus 4 is 3
-----------------------------------
-----------------------------------
THETAS  0.112 0.000 0.210
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
        1  -203.121527   -88.214370
-----------------------------------
TOTALS     -203.121527   -88.214370
-2 LN(LIKE) =  4.06243E+02
Maxcensor can be reduced to       -32767
-----------------------------------
-----------------------------------
THETAS  0.112 0.105 0.133
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
        1  -201.053315   -87.316159
-----------------------------------
TOTALS     -201.053315   -87.316159
-2 LN(LIKE) =  4.02107E+02
Maxcensor can be reduced to       -32767
-----------------------------------
-----------------------------------
THETAS  0.112 0.210 0.000
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
        1 -100000000000000000000.000000 -43429355638650388480.000000
-----------------------------------
TOTALS    -100000000000000000000.000000 -43429355638650388480.000000
-2 LN(LIKE) =  2.00000E+20
Maxcensor can be reduced to       -32767
***************************************************************************
Run   1 - LINKMAP : p12 p14 p1 p17
***************************************************************************
Program UNKNOWN version 5.1 (1-Feb-1991)
The following maximum values are in effect:
      10 loci
      55 single locus genotypes
      10 alleles at a single locus
     500 individuals in one pedigree
       3 marriage(s) for one male
       3 quantitative factor(s) at a single locus
      20 liability classes
      10 binary codes at a single locus
YOU ARE USING LINKAGE (V5.1 (1-Feb-1991)) WITH  4-POINT AUTOSOMAL DATA
Program LINKMAP version  5.10 (1-Feb-1991)
FASTLINK version  2.10 (21-Mar-1994)
The program constants are set to the following maxima:
     8 loci in mapping problem (maxlocus)
    12 alleles at a single locus (maxall)
   157 recombination probabilities (maxneed)
 50000 maximum of censoring array (maxcensor)
   180 haplotypes = n1 x n2 x ... where ni = current # alleles at locus i
 16290 joint genotypes for a female
 16290 joint genotypes for a male
  1000 individuals in all pedigrees combined (maxind)
 16290 joint genotypes for a male
  1000 individuals in all pedigrees combined (maxind)
   150 pedigrees (maxped)
    10 binary codes at a single locus (maxfact)
     3 quantitative factor(s) at a single locus
    20 liability classes
    10 binary codes at a single locus
    2.00 base scaling factor for likelihood (scale)
    2.00 scale multiplier for each locus (scalemult)
 0.00000 frequency for elimination of heterozygotes (minfreq)
YOU ARE USING LINKAGE (V5.1 (1-Feb-1991)) WITH  4-POINT
YOU ARE USING FASTLINK (V2.1 (21-Mar-1994)) AUTOSOMAL DATA
Number of alleles at locus 1 is 6
Number of alleles at locus 2 is 5
Number of alleles at locus 3 is 2
Number of alleles at locus 4 is 3
-----------------------------------
Number of alleles at locus 4 is 3
-----------------------------------
-----------------------------------
THETAS  0.112 0.000 0.210
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
        1  -203.121527   -88.214370
-----------------------------------
TOTALS     -203.121527   -88.214370
-2 LN(LIKE) =  4.06243E+02
Maxcensor can be reduced to       -32767
-----------------------------------
-----------------------------------
THETAS  0.112 0.070 0.163
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
        1  -201.247913   -87.400672
-----------------------------------
        1  -201.247913   -87.400672
-----------------------------------
TOTALS     -201.247913   -87.400672
-2 LN(LIKE) =  4.02496E+02
Maxcensor can be reduced to       -32767
-----------------------------------
-----------------------------------
THETAS  0.112 0.140 0.097
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
        1  -201.272873   -87.411512
-----------------------------------
TOTALS     -201.272873   -87.411512
-2 LN(LIKE) =  4.02546E+02
-----------------------------------
-----------------------------------
THETAS  0.112 0.210 0.000
-----------------------------------
THETAS  0.112 0.210 0.000
-----------------------------------
PEDIGREE |  LN LIKE  | LOG 10 LIKE
-----------------------------------
        1  -305.564464  -132.704678
-----------------------------------
TOTALS     -305.564464  -132.704678
-2 LN(LIKE) =  6.11129E+02



More information about the Gen-link mailing list