1. Tracking the genetic legacy of past human populations through the grid
Nicolas Ray
University of Bern / CMPG
(University of Geneva & UNEP/GRID-Europe)
ECSAC09, Veli Lošinj, August 26th 2009

2. Archaeology:
Paleoanthropology:

3. Human migrations Homo sapiens sapiens [12,000] [55,000]
Adapted from Cavalli-Sforza & Feldman, 2003
[12,000]
[55,000]
Let me set the scene of my talk
Homo sapiens sapiens

4. Why aiming at a good demographic model?

5. 1. Better understand human evolution
Origin of modern human (when, where, how many?)
Relationship with other members of the Homo genus
2. Distinguish between the effect of demography and those of selection (biomedical applications)
Distinguish between the effect of demography and those of selection
Find genes under selection
Exclude false positives

6. Observed patterns of genetic diversity in contemporary populations
A complex past demography
fluctuation in effective pop. size
substructure
migrations
Observed patterns of genetic diversity in contemporary populations
Gene-specific factors
mutations
recombination
selection

7. Semi-spatial approach

8. Statistical Evaluation of Alternative Models of Human Evolution
Nelson Fagundes, Nicolas Ray, Mark Beaumont, Samuel Neuenschwander, Francisco Salzano, Sandro Bonatto, and Laurent Excoffier PNAS, 104:
50 loci in non-genic regions (Chen and Li, 2001)
About 500 bp each, 24,425 bp in total
30 individuals: 10 Africans, 8 Asians, 12 Amerindians
Chimpanzee sequenced to get estimation of mutation rates assuming 6 My divergence time

9. Models African replacement Assimilation Multiregional evolution timeAM
AFRIG
AFREG AF AS AM
ASIG AF AS AM
ASEG AF AS AM
time
Multiregional evolution
MRE1S AF AS AM
MRE2S AF AS AM
MREBIG AF AS AM
MREBEG AF AS AM
There are two things that we are interested to know:
What is the most likely model
What are the likely parameter values (main two ones: exit of Africa, entrance into America)
 The statistical aprroach

10. Model parameters and priors

11. Simulations Coalescence theory
Africa
Asia
Americas
Time
A retrospective model of population genetics
Traces all copies of a gene in a sample from a population to a single ancestral copy shared by all members (MRCA)
Assumes no recombination, no selection
In genetics, coalescent theory is a retrospective model of population genetics that traces all alleles of a gene in a sample from a population to a single ancestral copy shared by all members of the population, known as the most recent common ancestor (MRCA; sometimes also termed the coancestor to emphasize the coalescent relationship[1]). The inheritance relationships between alleles are typically represented as a gene genealogy, similar in form to a phylogenetic tree. This gene genealogy is also known as the coalescent; understanding the statistical properties of the coalescent under different assumptions forms the basis of coalescent theory. In the most simple case, coalescent theory assumes no recombination, no natural selection, and no gene flow or population structure. Advances in coalescent theory, however, allow extension to the basic coalescent, and can include recombination, selection, and virtually any arbitrarily complex evolutionary or demographic model in population genetic analysis. The mathematical theory of the coalescent was originally developed in the early 1980s by John Kingman[2]

12. Simulated genealogy Summary statistics Within population: S, p
TCCTTGTA…ATTGGT
Mutation
Modèle de mutation
ACCTAGTACAATCGGTAATGCCATTGGT
Summary statistics
Within population:
S, p
Between populations
Pairwise FST
Global FST
Globally
This process is repeated for each independent loci
ACCGAGTA…GTTGGT

13. Approximate Bayesian Computations (ABC)
The rejection-sampling approach:
Calculate summary statistics (S) for observed data sets
Draw parameter values φ’ from prior distributions, and use them to simulate data
Calculate summary statistics (S’) on the simulated data set and compare them to the observations: δ = ||S - S’|| (Euclidean distance)
Accept φ’ if δ is arbitrarily small, otherwise reject sample
Rejection-sampling method
Tavaré et al. 1997; Pritchard et al. 1999
Aim:
Estimate some parameters F in a well defined model
The ABC approach (Beaumont et al. 2002)
Modification: a local regression is added within the set of accepted φ’ values

14. Rejection-sampling method Tavaré et al. 1997; Pritchard et al. 1999
Aim:
Estimate some parameters F in a well defined model
Neuenschwander (2006)

15. Computational issues Computer clusters UBELIX (>500 nodes)
Draw parameter values from priors
Simulate one genealogy
Generate genetic data
Compute summary statistics
Computer clusters
UBELIX (>500 nodes)
Zooblythii (~40 nodes)
1-10 mio.

16. For ABC, 5 mio. demographic simulations are necessary to obtain robust parameter estimations
Each demographic simulation is followed by n genetic simulations (n = num. of loci)
Example
8 simple models, 50 loci, 30 individuals  2 CPU-year

17. Relative probabilities of models of human evolution
African replacement
Assimilation
AFREG AF AS AM AF AS AM
AFRIG
AFREG AF AS AM
ASEG AF AS AM
ASIG AF AS AM
ASEG AF AS AM
0.781
0.001
0.958
0.042
0.091
0.909
Multiregional evolution
MREBIG AF AS AM
MRE1S AF AS AM
MRE2S AF AS AM
MREBIG AF AS AM
MREBEG AF AS AM
0.218
0.461
0.422
0.048
0.069

18. Americas colonization time
51.1 Kya
(40.1 – 70.9)
Out-of-Africa time
AFREG AS AM AF
10.3 Kya
(7.6 – 15.9)
Americas colonization time
142 Kya
(104 – 186)
Speciation time

19. Fully spatial approach

20. A complex demography [10,000] demographic and spatial expansions
Adapted from Cavalli-Sforza & Feldman, 2003
[10,000]
demographic and spatial expansions
[55,000]
population bottlenecks
secondary contacts
! Show landscape effects on demography: bottlenecks, isolation, expansion, migration, secondary contact
Now, it seems on this figure that the environment of the world is quite uniform, but of course it’s not the case…
population isolation
fast migration events

21. From environment to demography
low
high
Carrying
capacity
Basée sur les données de la littérature sur les densités observées des chasseurs-cueilleurs contemporains.
Vous remarquez que la projection géographique a été modifiée, afin de minimiser les distorsions.
Spatial resolution: 100 km

22. From environment to demography
Friction
low
Basée sur quelques données de la littérature, mais encore moins de données que K
high

23. Demographic simulations
stepping-stone model (cellular automata)
Cell or deme
Pop. size
time

24. SPatiaL And Temporal Coalescences in Heterogeneous Environment
SPLATCHE
SPatiaL And Temporal Coalescences in Heterogeneous Environment (
Windows-based graphical version: extremely useful to explore outputs of simulation
Of course a Console-version for faster execution of simulation under Linux-based system

25. Vegetation maps Vegetation at the Last Glacial Maximum
Taking into account altitudes
Expert system
present potential
Dvpt avec Jonathan Adams
Ray et Adams Internet Archaeology 11
Vegetation at the Last Glacial Maximum
Present potential vegetation
Last Glacial Maximum

26. Demography and spatial expansion
Population density
Ce n’est qu’un exemple

27. Dynamic vegetation
intermediate PP
LGM
Ce n’est qu’un exemple

29. Genetic simulations
we need to sample indigenous people, and avoid admixed people

30. Computational issues
A fully spatially-explicit model using 500 loci in 800 individuals:
 10 CPU-years
Adding long-distance dispersal:
 20 CPU-years

31. SPLATCHE on the grid
early 2005: joined the Biomed VO of the EGEE project
mid 2005: tested on GILDA test bed, and deployed on the Grid
since late 2005: testing and improvement
since mid 2006: production mode and optimization

32. Use of SPLATCHE on the grid
N simulations
Posterior distribution of demographic/genetic parameters of interest
Statistical tools
Independent simulations:
the more CPUs, the better
job failures are not that bad

33. Optimizations Reduction of the number of simulations (Daniel Wegmann)
Submission time
multi-threaded application using up to 30 RBs (used for the WISDOM project)
Fetching time of job outputs
in-house multi-threaded solution for checking status and getting outputs
GRID
5 mio. simulations
4’000 jobs takes about 12 hours with sequential submission
WISDOM is a Initiative for grid-enabled drug discovery against neglected and emergent diseases
Wide In Silico Docking On Malaria
I thank the developers of WISDOM, the team of Vincent Breton, and especially Mathieu Reichstadt (?) et Jean Weismann.
Reduction of the number of simulations (Daniel Wegmann)
By MCMC. Promising results (~10 times less sims)

34. African replacement (AFREG) model best fits with observed genetic diversity in non-genic regions
Scenarios with interbreeding are less well supported
Colonization of Americas seems to have occurred post LGM
Use of additional statistics, loci, samples may increase the power of the method

35. Geographic origin of human dispersal
22 populations
Talk briefly about multiregional evolution scenarios
Ray et al. (2005) Genome Research

37. Mutations surfing during a range expansion

38. Mutations surfing during a range expansion
Klopfstein, Currat and Excoffier (2006) MBE 23(3):
Some mutation can travel with the wave of advance
New mutations can reach high frequencies
More pronounced in small populations
Spatial distribution of the frequency of a new mutation. K=10
Phenomenon inferred from simulations, difficult to study analytically
Centroid of the spatial distribution is often far from the origin

39. Selection ?
Currat, Excoffier, Maddison, Otto, Ray, Whitlock and Yeaman (2006) Science 313:172a
La découverte du surfing souligne l'importance d'utiliser des modèles réalistes pour l'évolution humaines avant de tirer des conclusions définitives sur la sélection de certains allèles
Microcephalin: about 60% difference bw Africa and the rest (for a particular haplogroup)
ASPM: about 20% difference bw Africa and the rest
ASPM : (abnormal spindle-like microcephaly associated)
(2005) Science 509 (5741)

40. Interactions among populations
Interaction between modern humans and Neanderthals in Europe
Currat & Excoffier (2004), PLoS Biol.

41. Cane toad invasion in Australia
Estoup, A., Baird, S. J. E., Ray, N., Currat, M., Cornuet, J.-M., Santos, F., Beaumont, M. A. and L. Excoffier. Combining genetic, historical and geographic data to reconstruct the dynamics of the bioinvasion of cane toad Bufo marinus. In prep

42. Take-home message A good human demographic model is important
Realistic spatially-explicit approaches are essential
The grid is key for sufficient exploration of parameter space
User support and connections outside one’s discipline is crucial

43. Thank you!