12-minute presentation given at SMBE 2015 in Vienna, in the PopGen in Space! symposium.
Related paper coming out in September 2015 in The American Naturalist.
Ride the Storm: Navigating Through Unstable Periods / Katerina Rudko (Belka G...
Evolution of quantitative traits under a migration–selection balance: When does skew matter?
1. Evolution of quantitative traits
under a migration–selection balance:
When does skew matter?
S. Yeaman,F. Débarre, F. Guillaume.
@flodebarre @sam_yeaman @fred_guillaume
Débarre, Yeaman & Guillaume Skew matters? SMBE – -- /
7. Divergent selection and trait distributions
Migration
Optimum ✓ Optimum ✓
Traits
0
Traits
0
?Are the local distributions Gaussian?
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
8. Divergent selection and trait distributions
Migration
Optimum ✓ Optimum ✓
Traits
0
Traits
0
?Are the local distributions Gaussian?
If not, can we nevertheless approximate them as Gaussian
to predict phenotypic divergence between demes?
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
9. Outline
Analytics: include the eect of skew in predictions
of phenotypic divergence;
Simulations: estimate the amount of skew for
dierent genetic architectures.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
10. Outline
Analytics: include the eect of skew in predictions
of phenotypic divergence;
Simulations: estimate the amount of skew for
dierent genetic architectures.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
18. The eect of skew on phenotypic dierentiation
D⇤
= z⇤
z⇤
.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
19. The eect of skew on phenotypic dierentiation
D⇤
= z⇤
z⇤
.
Without specifying the distribution
D⇤
=
vm⇤ ✓ + m⇤
vm⇤ ( m) + m !
(with m⇤
).
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
20. The eect of skew on phenotypic dierentiation
D⇤
= z⇤
z⇤
.
Without specifying the distribution
D⇤
=
vm⇤ ✓ + m⇤
vm⇤ ( m) + m !
(with m⇤
).
Assuming a Gaussian distribution, and constant variance
D⇤
G =
v⇤ ✓
v⇤ ( m) + m !
.
(Hendry et al., )
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
21. The eect of skew on phenotypic dierentiation
D⇤
= z⇤
z⇤
.
Without specifying the distribution
D⇤
=
vm⇤ ✓ + m⇤
vm⇤ ( m) + m !
(with m⇤
).
Assuming a Gaussian distribution, and constant variance
D⇤
G =
v⇤ ✓
v⇤ ( m) + m !
.
(Hendry et al., )
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
22. The eect of skew on phenotypic dierentiation
D⇤
= z⇤
z⇤
.
Without specifying the distribution
D⇤
=
vm⇤ ✓ + m⇤
vm⇤ ( m) + m !
(with m⇤
).
Assuming a Gaussian distribution, and constant variance
D⇤
G =
v⇤ ✓
v⇤ ( m) + m !
.
(Hendry et al., )
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
23. Outline
Analytics: include the eect of skew in predictions
of phenotypic divergence;
Simulations: estimate the amount of skew for
dierent genetic architectures.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
24. Outline
Analytics: include the eect of skew in predictions
of phenotypic divergence;
Simulations: estimate the amount of skew for
dierent genetic architectures.
a Diallelic model, influence of eect size;
b Continuum of alleles model, evolvable eects.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
25. Outline
Analytics: include the eect of skew in predictions
of phenotypic divergence;
Simulations: estimate the amount of skew for
dierent genetic architectures.
a Diallelic model, influence of eect size;
b Continuum of alleles model, evolvable eects.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
26. Diallelic model
unlinked loci
alleles at each locus: + and .
loci of eect .,
locus of eect a.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
27. Diallelic model: Results
a = .
m = ⇤
−Θ 0 Θ
Trait
Density
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
28. Diallelic model: Results
a = .
m = ⇤
−Θ 0 Θ
Trait
Density
a = .
m = ⇤
−Θ 0 Θ
Trait
Density
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
29. Diallelic model: Results
a = .
m = ⇤
−Θ 0 Θ
Trait
Density
0
∆Θ 2
∆Θ
10−6
10−5
10−4
10−3
10−2
10−1
Migration m
DifferentiationD*
D⇤
, D⇤
G, simulations.
a = .
m = ⇤
−Θ 0 Θ
Trait
Density
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
30. Diallelic model: Results
a = .
m = ⇤
−Θ 0 Θ
Trait
Density
0
∆Θ 2
∆Θ
10−6
10−5
10−4
10−3
10−2
10−1
Migration m
DifferentiationD*
D⇤
, D⇤
G, simulations.
a = .
m = ⇤
−Θ 0 Θ
Trait
Density
0
∆Θ 2
∆Θ
10−6
10−5
10−4
10−3
10−2
10−1
Migration m
DifferentiationD*
D⇤
, D⇤
G, simulations.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
31. Diallelic model: Results
a = .
m = ⇤
−Θ 0 Θ
Trait
Density
0
∆Θ 2
∆Θ
10−6
10−5
10−4
10−3
10−2
10−1
Migration m
DifferentiationD*
D⇤
, D⇤
G, simulations.
a = .
m = ⇤
−Θ 0 Θ
Trait
Density
0
∆Θ 2
∆Θ
10−6
10−5
10−4
10−3
10−2
10−1
Migration m
DifferentiationD*
D⇤
, D⇤
G, simulations. ⇤
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
36. Outline
Analytics: include the eect of skew in predictions
of phenotypic divergence;
Simulations: estimate the amount of skew for
dierent genetic architectures.
a Diallelic model, influence of eect size;
b Continuum of alleles model, evolvable eects.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
37. Continuum-of-alleles model
unlinked loci
Continuum of alleles at each locus.
Mutation: z0
(i) = z(i) + dz,
with dz normally distributed (N(, ↵)).
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
43. Summary // Take-Home messages
I Neglecting the skew of distributions can lead to the
underestimation of phenotypic divergence D⇤
D⇤
=
vm⇤
✓ + m⇤
vm⇤ ( m) + m !
;
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
44. Summary // Take-Home messages
I Neglecting the skew of distributions can lead to the
underestimation of phenotypic divergence D⇤
D⇤
=
vm⇤
✓ + m⇤
vm⇤ ( m) + m !
;
I There is virtually no eect of the timing of measurement of
variance and skew (e.g., before/aer dispersal) on D⇤;
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
45. Summary // Take-Home messages
I Neglecting the skew of distributions can lead to the
underestimation of phenotypic divergence D⇤
D⇤
=
vm⇤
✓ + m⇤
vm⇤ ( m) + m !
;
I There is virtually no eect of the timing of measurement of
variance and skew (e.g., before/aer dispersal) on D⇤;
I Skew is more likely to be generated when there are loci of
major eect;
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
46. Summary // Take-Home messages
I Neglecting the skew of distributions can lead to the
underestimation of phenotypic divergence D⇤
D⇤
=
vm⇤
✓ + m⇤
vm⇤ ( m) + m !
;
I There is virtually no eect of the timing of measurement of
variance and skew (e.g., before/aer dispersal) on D⇤;
I Skew is more likely to be generated when there are loci of
major eect;
I But under the conditions of the infinitesimal model, little skew
is generated: the Gaussian approximation remains a good one.
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
47. Summary // Take-Home messages
I Neglecting the skew of distributions can lead to the
underestimation of phenotypic divergence D⇤
D⇤
=
vm⇤
✓ + m⇤
vm⇤ ( m) + m !
;
I There is virtually no eect of the timing of measurement of
variance and skew (e.g., before/aer dispersal) on D⇤;
I Skew is more likely to be generated when there are loci of
major eect;
I But under the conditions of the infinitesimal model, little skew
is generated: the Gaussian approximation remains a good one.
I For more details, see
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
48. Summary // Take-Home messages
I Neglecting the skew of distributions can lead to the
underestimation of phenotypic divergence D⇤
D⇤
=
vm⇤
✓ + m⇤
vm⇤ ( m) + m !
;
I There is virtually no eect of the timing of measurement of
variance and skew (e.g., before/aer dispersal) on D⇤;
I Skew is more likely to be generated when there are loci of
major eect;
I But under the conditions of the infinitesimal model, little skew
is generated: the Gaussian approximation remains a good one.
I For more details, see
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /
49. Summary // Take-Home messages
I Neglecting the skew of distributions can lead to the
underestimation of phenotypic divergence D⇤
D⇤
=
vm⇤
✓ + m⇤
vm⇤ ( m) + m !
;
I There is virtually no eect of the timing of measurement of
variance and skew (e.g., before/aer dispersal) on D⇤;
I Skew is more likely to be generated when there are loci of
major eect;
I But under the conditions of the infinitesimal model, little skew
is generated: the Gaussian approximation remains a good one.
I For more details, see
Thanks for your attention!
Débarre, Yeaman Guillaume Skew matters? SMBE – -- /