Biometrics and Statistics Unit
Francisco Manuel Rodríguez Huerta ([email protected])
Centro Internacional de Mejoramiento de Maíz y Trigo, Internacional
1 Basic model 2 North Carolina NCI NCII 3 Line by Tester 4 Gring Diallel Methods Models
Basic genetic model.
yik = µ + sk+ gi+ eik
Interaction model.
yik = µ + sk + gi + sk∗ gi + eik
yijk is the observed value
µ is the mean
sk is the non random eect (k = 1, 2, ..., s)
gi is the genotype eect (i = 1, 2, ..., g)
North Carolina I
In this type of design due to Comstock and Robinson (1952), a number m of males are selected from an F2 gen-eration or an advanced gengen-eration by random mating, and each of the se-lected males are crossed with two or more f dierent females. This gives mf full sib (FS) families.
Balanced: the number of crosses is m ∗ f crosses, in the example f = 4 Unbalanced: the number of crosses is lower than m ∗ f , example: no in-formation for cross 1x2
RCBD can be modeled using:
yijk= µ + REPk + gij+ eij
= µ + REPk + mi + fj (i )+ eijk
yijk is the observed value
µ is the mean
REPk is the replicate eect (k = 1, 2, ..., r)
mi is the male eect (i = 1, 2, ..., m)
fj (i ) is the eect of female j in male i (j = 1, 2, ..., f )
North Carolina I
Alpha Lattice can be modeled using:
yijk = µ + REPk + gij + BLK (REPk) + eij
= µ + REPk + mi + fj (i )+ BLK (REPk) + eijk
yijk is the observed value
µ is the mean
REPk the replicate eect (k = 1, 2, ..., r)
mi is the male eect (i = 1, 2, ..., m)
fj (i ) is the eect of female j in male i (j = 1, 2, ..., f )
BLK (REPk) random eect of block nested in replicate k
VC Henderson (balanced) REML (Balanced or unbalanced) σm2 (MSm− MSf (m))/rm σˆ2m σf (m)2 (MSf (m)− MSe)/r σˆf (m)2 σA2 4σ2m σ2D 4(σf (m)2 − σm2) σ2E σe2 r hb2 σ2A+σD2 σ2A+σD2+σ2E hn2 σA2 σ2A+σD2+σ2E
North Carolina I
Multi-environment RCBD (interaction model) can be modeled using:
yijkd = µ + Ed+ REPk(Ed) + gij + Ed∗ gij + eijkd
= µ + Ed+ REPk(Ed) + mi+ fj (i )+ Ed∗ mi+ Ed∗ fj (i )+ eijkd
yijkd is the observed value
µ is the mean
Ed environmental eect (d = 1, 2, ..., s)
REPk(Ed)the eect of replicate k nested in environment d
(k = 1, 2, ..., r)
mi is the male eect (i = 1, 2, ..., m)
fj (i ) is the eect of female j in male i (j = 1, 2, ..., f )
Multi-environment Alpha Lattice can be modeled using:
yijkd = µ + Ed+ REPk(Ed) + gij + Ed∗ gij + BLK (REPkEd) + eijkd
= µ + Ed+ REPk(Ed) + mi+ fj (i )+ Ed∗ mi+ Ed∗ fj (i )
+ BLK (REPkEd) + eijkd
yijkd is the observed value
µ is the mean
Ed environmental eect (d = 1, 2, ..., s)
REPk(Ed)the eect of replicate k nested in environment d
(k = 1, 2, ..., r)
mi is the male eect (i = 1, 2, ..., m)
fj (i ) is the eect of female j in male i (j = 1, 2, ..., f ) BLK (REPkEn) random eect of block nested in replicate k
nested in environment d eijkd residual.
North Carolina I
VC Henderson (balanced) REML (Balanced or unbalanced)
σm2 (MSm− MSf (m))/srm σˆ2m σ2f (m) (MSf (m)− MSe)/sr σˆ2f (m) σ2A 4σ2m σD2 4(σf (m)2 − σm2) σ2E σ2gxe s + σ2e sr h2b σ2A+σD2 σ2A+σD2+σ2E h2n σA2 σ2A+σD2+σ2E
In this design (Comstock and Robinson (1952)), a set of males m and a set of females f are selected, say from an F2 population. Each of the males is crossed with every female in the set, and thus mf FS families are generated.
Balanced: the number of crosses is m ∗ f crosses
Unbalanced: the number of crosses is lower than m ∗ f , example: no information for cross 1x2
Male\Female 1 2 . . . f
1 x11 x12 . . . x1f
2 x21 x22 . . . x2f
... ... ... ... ...
North Carolina II
RCBD can be modeled using:
yijk = µ + REPk+ gij + eij
= µ + REPk+ mi+ fj + mi ∗ fj + eijk
yijk is the observed value
µ is the mean
REPk the replicate eect (k = 1, 2, ..., r)
mi is the male eect (i = 1, 2, ..., m)
fj is the female eect (j = 1, 2, ..., f )
Alpha Lattica can be modeled using: yijk = µ + REPk+ gij + BLK (REPk)eij
= µ + REPk+ mi + fj + mi∗ fj + BLK (REPk) + eijk
yijk is the observed value
µ is the mean
REPk the replicate eect (k = 1, 2, ..., r)
mi is the male eect (i = 1, 2, ..., m)
fj is the female eect (j = 1, 2, ..., f )
BLK (REPk) random eect of block nested in rep
North Carolina II
VC Henderson (balanced) REML (Balanced or unbalanced)
σm2 (MSm− MSmf)/rm σˆ2m σ2f (MSf − MSmf)/rf σˆ2f σ2mf (MSmf − MSe)/r σˆ2mf σ2g (m−1)MSm+(f −1)MSf−(m+f −2)MSmf r (2mf −m−f ) σˆ2g σA2 4σg2 σD2 4σmf2 σ2E σ2e/r hb2 σA2+σ2D σA2+σ2D+σE2 hn2 σA2 σA2+σ2D+σE2
Multi-environment RCBD can be modeled using:
yijkd = µ + Ed+ REPk(Ed) + gij + Ed∗ gij + eijkd
= µ + Ed+ REPk(Ed) + mi+ fj + mi ∗ fj
+ Ed∗ mi+ Ed∗ fj + Ed ∗ mi ∗ fj + eijkd
yijkd is the observed value
µ is the mean
Ed environmental eect (d = 1, 2, ..., s)
REPk(Ed)the eect of replicate k nested in environment d
(k = 1, 2, ..., r)
mi is the male eect (i = 1, 2, ..., m)
fj is the female eect (j = 1, 2, ..., f )
North Carolina II
Multi-environment Alpha Lattica can be modeled using:
yijkd = µ + Ed+ REPk(Ed) + gij + S ∗ gij + BLK (REPkEd) + eijkd
= µ + Ed+ REPk(Ed) + mi+ fj+ mi∗ fj
+ Ed∗ mi+ Ed ∗ fj + Ed∗ mi ∗ fj + BLK (REPkEd) + eijkd
yijkd is the observed value
µ is the mean
Ed environmental eect (d = 1, 2, ..., s)
REPk(Ed)the eect of replicate k nested in environment d
(k = 1, 2, ..., r)
mi is the male eect (i = 1, 2, ..., m)
fj is the female eect (j = 1, 2, ..., f )
BLK (REPkEd) random eect of block nested in replicate k
nested in environment d eijkd residual.
VC Henderson (balanced) REML (Balanced or unbalanced) σm2 (MSm− MSmf)/srm σˆ2m σ2f (MSf − MSmf)/srf σˆ2f σ2mf (MSmf − MSe)/sr σˆ2mf σ2g (m−1)MSm+(f −1)MSf−(m+f −2)MSmf sr (2mf −m−f ) σˆ2g σA2 4σg2 σD2 4σmf2 σ2E σ2gxe s + σ2e sr hb2 σA2+σ2D σA2+σ2D+σE2 hn2 σA2 σA2+σ2D+σE2
Line by Tester
Kempthorne (1957) proposed a line Ö tester experiment, with similar mating pattern as the North Carolina design II.
Balanced: the number of crosses is l ∗ t crosses, in the example
f =4
Unbalanced: the number of crosses is lower than l ∗ t , example: no information for cross 1x2
Line\Tester 1 2 . . . t
1 x11 x12 . . . x1t
2 x21 x22 . . . x2t
... ... ... ... ...
RCBD can be modeled using:
yijk = µ + REPk+ li+ tj+ li∗ tj + eijk
yijk is the observed value
µ is the mean
REPk the replicate eect (k = 1, 2, ..., r)
li is the male eect (i = 1, 2, ..., l)
tj is the female eect (j = 1, 2, ..., t)
Line by Tester
Alpha Latice can be modeled using:
yijk = µ + REPk+ li + tj + li ∗ tj + BLK (REPk) + eijk
yijk is the observed value
µ is the mean
REPk is the replicate eect (k = 1, 2, ..., r)
li is the male eect (i = 1, 2, ..., l)
tj is the female eect (j = 1, 2, ..., t)
BLK (REPk) random eect of block nested in replicate k
VC Henderson (balanced) REML (Balanced or unbalanced) σl2 (MSl− MSlt)/rt σˆ2l σt2 (MSt− MSlt)/rl σˆ2t σ2lt (MSlt− MSe)/r σˆlt2 σg2 (l −1)MSl+(t−1)MSt−(l+t−2)MSlt r (2lt−l−t) σˆ2g σ2A 4σg2 σD2 4σlt2 σ2E σ2e r h2b σA2+σ2D σA2+σ2D+σ2E h2n σ2A σA2+σ2D+σ2E
Line by Tester
Multi-environment RCBD can be modeled using:
yijkd = µ + Ed+ REPk(Ed) + gij + Ed∗ gij + eijkd
= µ + Ed+ REPk(Ed) + li+ tj+ li∗ tj
+ Ed∗ li+ Ed∗ tj + Ed∗ li ∗ tj + eijkd
yijkd is the observed value
µ is the mean
Ed environmental eect (d = 1, 2, ..., s)
REPk(Ed)the eect of replicate k nested in environment d
(k = 1, 2, ..., r)
li is the male eect (i = 1, 2, ..., l)
tj is the female eect (j = 1, 2, ..., t)
Multi-Environment Alpha Latice can be modeled using:
yijkd = µ + Ed+ REPk(Ed) + gij + Ed∗ gij + BLK (REPkEd) + eijkd
= µ + Ed+ REPk(Ed) + li+ tj + li∗ tj
+ Ed∗ li+ Ed∗ tj + Ed∗ li ∗ tj + BLK (REPkEd) + eijkd,
yijkd is the observed value
µ is the mean
Ed environmental eect (d = 1, 2, ..., s)
REPk(Ed)the eect of replicate k nested in environment d
(k = 1, 2, ..., r)
li is the male eect (i = 1, 2, ..., l)
tj is the female eect (j = 1, 2, ...t)
BLK (REPkEd) random eect of block nested in replicate k
nested in environment d eijkd residual.
Line by Tester
VC Henderson (balanced) REML (Balanced or unbalanced)
σl2 (MSl− MSlt)/srt σˆ2l σt2 (MSt− MSlt)/srl σˆ2t σ2lt (MSlt− MSe)/sr σˆlt2 σg2 (l −1)MSl+(t−1)MSt−(l+t−2)MSlt sr (2lt−l−t) σˆ2g σ2A 4σg2 σD2 4σlt2 σ2E σ2gxe s + σe2 sr h2b σA2+σ2D σA2+σ2D+σ2E h2n σ2A σA2+σ2D+σ2E
Gring (1956) considered four methods for generating the crosses among a set of p inbred lines and their analysis for comparative and exploratory experiments. Gring presented four mating designs, called methods, and four models by taking the genetic eects as xed or random.
Methods
Diallel crossing techniques may vary depending upon whether or not the parental inbreeds or the reciprocal F0
1s are included or both.
With this as a basis for classication there are four diallel crossing systems or experimental methods:
a) Method 1: Parents, one set of F0
1s and reciprocal F10s are
included (all p2 combinations) b) Method 2: Parents and one set of F0
1s are included but
reciprocal F0
1s are not (p(p + 1)/2 combinations) c) Method 3: One set of F0
1s and reciprocal F10s are included but
not the parents (p(p − 1) combinations)
d) Method 4: One set of F0
1s but neither parents nor reciprocal
a) Parents 1 2 3 1 x11 x12 x13 2 x21 x22 x23 3 x31 x32 x33 b) Parents 1 2 3 1 x11 x12 x13 2 x22 x23 3 x33 c) Parents 1 2 3 1 x12 x13 2 x21 x23 3 x31 x32 d) Parents 1 2 3 1 x12 x13 2 x23 3
Models
Model 1: the genotype and block eects are constants Model 2: the genotype and block eects are both random variables
Model A (mixed A): the genotype eects are random variables and the block eects are constants
Model B (mixed B): the genotype eects are constants and block eects are random variables
RCBD or Alpha Lattice can be modeled respectively using: yijk = µ + REPk+ gij + eijk
= µ + REPk+ gcai + scaij + mi+ rij + eijk
or
yijk= µ + REPk + gcai+ scaij + mi + rij + BLK (REPk) + eijk
yijk is the observed value
µ is the mean
REPk is the replicate eect (k = 1, 2, ..., r)
gcai is the general combining ability eect (i = 1, 2, ..., p)
scaij is the specic combining ability eect (i = 1, 2, ..., p and
j =1, 2, ..., p)
mi is the maternal eect (i = 1, 2, ..., p)
rij is the reciprocal eect (i = 1, 2, ..., p and j = 1, 2, ..., p)
BLK (REPk) random eect of block nested in replicate k
Variance components Method 1
VC model 1 and mixed B model 2 and mixed A σgca2 2rp1 (MSgca− MSe) 2rp1 MSgca− MSe−p(p−1)MSsca (p(p−1)+1) σ2sca (MSsca− MSe)/r p 2 2r(p(p−1)+1)(MSsca− MSe) σm2 (MSm− MSr)/2pr (MSm− MSr)/2pr σ2r (MSr− MSe)/2r (MSr− MSe)/2r σ2ρ 2σgca2 + σ2sca+2σm2 + σ2r +σ2e r σA2 2σ2gca σ2D σ2sca hb2 σA2+σ2D σρ2 h2n σA2 σρ2
VC model 1 and mixed B model 2 and mixed A σgca2 2r(p−2)1 (MSgca− MSe) 2r(p−2)1 (MSgca− MSsca)
σ2sca (MSsca− MSe)/2r (MSsca− MSe)/2r
σ2m (MSm− MSr)/2pr (MSm− MSr)/2pr σ2r (MSr− MSe)/2r (MSr− MSe)/2r σ2ρ 2σgca2 + σ2sca+2σm2 + σ2r +σ2e r σA2 2σ2gca σ2D σ2sca h2b σA2+σ2D σρ2 h2n σA2 σρ2
Models for Method 1 and Method 3 Multi-environment
RCBD or Alpha Lattice can be modeled respectively using: yijkd = µ + Ed+ REPk(Ed) + gij + Ed∗ gij + eijkd
= µ + Ed+ REPk(Ed) + gcai + scaij + mi+ rij
+ Ed∗ gcai + Ed∗ scaij + Ed∗ mi + Ed∗ rij + eijkd
or
yijkd = µ + Ed+ REPk(Ed) + gcai+ scaij + mi + rij
+ Ed∗ gcai+ Ed ∗ scaij + Ed∗ mi+ Ed∗ rij
+ BLK (REPkEd) + eijkd
Ed environmental eect (d = 1, 2, ..., s)
REPk(Ed)the eect of replicate k nested in environment d
(k = 1, 2, ..., r)
BLK (REPkEd) random eect of block nested in replicate k
VC model 1 and mixed B model 2 and mixed A σ2gca 2srp1 (MSgca− MSe) 2srp1 MSgca− MSe−p(p−1)MSsca (p(p−1)+1) σsca2 sr1(MSsca− MSe) p 2 2sr(p(p−1)+1)(MSsca− MSe) σm2 2srp1 (MSm− MSr) 2srp1 (MSm− MSr) σ2r 2sr1 (MSr− MSe) 2sr1 (MSr− MSe)
σ2ρ 2σ2gca+ σsca2 +2σ2m+ σr2+σgxe2
s + σ2e sr σA2 2σ2gca σD2 σ2sca hb2 σA2+σ2D σρ2 hn2 σA2 σρ2
Variance components Method 3 Multi-environment
VC model 1 and mixed B model 2 and mixed A σ2gca 2sr(p−2)1 (MSgca− MSe) 2sr(p−2)1 (MSgca− MSsca)
σsca2 2sr1 (MSsca− MSe) 2sr1 (MSsca− MSe)
σm2 2srp1 (MSm− MSr) 2srp1 (MSm− MSr) σ2r 2sr1 (MSr− MSe) 2sr1 (MSr − MSe) σ2ρ 2σ2gca+ σsca2 +2σ2m+ σr2+σ 2 gxe s + σ2e sr σA2 2σ2gca σD2 σ2sca hb2 σA2+σ2D σρ2 hn2 σA2 σρ2
RCBD or Alpha Lattice can be modeled respectively using: yijk = µ + REPk+ gij + eijk
= µ + REPk+ gcai+ scaij + eijk
or
yijk = µ + REPk + gcai+ scaij + BLK (REPk) + eijk
yijk is the observed value
µ is the mean
REPk is the replicate eect (k = 1, 2, ..., r)
gcai is the general combining ability eect (i = 1, 2, ..., p)
scaij is the specic combining ability eect (i = 1, 2, ..., p and
j =1, 2, ..., p)
BLK (REPk) random eect of block nested in replicate k
Variance components Method 2
VC model 1 and mixed B model 2 and mixed A σgca2 (p+12)r(MSgca− MSe) r (p+12)(MSgca− MSsca)
σ2sca 1r(MSsca− MSe) 1r(MSsca− MSe)
σ2ρ 2σgca2 + σ2sca+σ2e r σA2 2σ2gca σ2D σ2sca h2b σA2+σ2D σρ2 h2n σA2 σρ2
VC model 1 and mixed B model 2 and mixed A σ2gca (p−12)r(MSgca− MSe) r (p−12)(MSgca− MSsca)
σsca2 1r(MSsca− MSe) 1r(MSsca− MSe)
σ2ρ 2σgca2 + σ2sca+σ2e r σA2 2σ2gca σD2 σ2sca hb2 σA2+σ2D σρ2 hn2 σA2 σρ2
Models for Method 2 and Method 4 Multi-environment
RCBD or Alpha Lattice can be modeled respectively using: yijkd = µ + Ed+ REPk(Ed) + gij + Ed∗ gij + eijkd
= µ + REPk+ gcai + scaij
+ Ed∗ gcai + Ed∗ scaij + eijkd
or
yijkd = µ + Ed+ REPk(Ed) + gcai+ scaij
+ Ed∗ gcai+ Ed∗ scaij + BLK (REPkEd) + eijkd
Ed environmental eect (d = 1, 2, ..., s)
REPk(Ed)the eect of replicate k nested in environment d
(k = 1, 2, ..., r)
BLK (REPkEd) random eect of block nested in replicate k
VC model 1 and mixed B model 2 and mixed A σgca2 sr (p+1 2)(MSgca− MSe) sr (p+1 2)(MSgca− MSsca)
σ2sca sr1(MSsca− MSe) sr1(MSsca− MSe)
σρ2 2σ2gca+ σsca2 +2σ2m+ σr2+σgxe2
s + σ2e sr σ2A 2σ2gca σ2D σ2sca h2b σA2+σ2D σρ2 h2n σA2 σρ2
Variance components Method 4 Multi-environment
VC model 1 and mixed B model 2 and mixed A σ2gca sr (p−1 2)(MSgca− MSe) sr (p−1 2)(MSgca− MSsca)
σsca2 sr1(MSsca− MSe) sr1(MSsca− MSe)
σρ2 2σ2gca+ σsca2 +2σ2m+ σr2+σgxe2
s + σ2e sr σ2A 2σ2gca σD2 σ2sca h2b σA2+σ2D σρ2 h2n σA2 σρ2
H. Hinkelmann.
Design and Analysis Handbook of Experiments Volume 3 Special Designs and applications.
WILEY, 2012.
B. Ggring. 1956b
Concept of general and specic combining ability in relation to diallel crossing systems
Martínez, G. A.
Diseño y análisis de los experimentos de cruzas dialélicas.