Computational Insights and the Theory of Evolution

47
Computational Insights and the Theory of Evolution Christos H. Papadimitriou UC Berkeley

description

Computational Insights and the Theory of Evolution. Christos H. Papadimitriou UC Berkeley. Evolution Before Darwin. Erasmus Darwin. Before Darwin. J.-B. Lamarck. Before Darwin. Charles Babbage [Paraphrased] “ God created not species, but the Algorithm for creating species ”. - PowerPoint PPT Presentation

Transcript of Computational Insights and the Theory of Evolution

Page 1: Computational Insights and the Theory of Evolution

Computational Insightsand the Theory of Evolution

Christos H. Papadimitriou

UC Berkeley

Page 2: Computational Insights and the Theory of Evolution

Evolution Before Darwin

• Erasmus Darwin

Page 3: Computational Insights and the Theory of Evolution

Before Darwin

• J.-B. Lamarck

Page 4: Computational Insights and the Theory of Evolution

Before Darwin

• Charles Babbage

[Paraphrased]

“God created not species, but the Algorithm for creating species”

Page 5: Computational Insights and the Theory of Evolution

Darwin, 1858

•Common Ancestry•Natural Selection

Page 6: Computational Insights and the Theory of Evolution

The Origin of Species

• Possibly the world’s most masterfully compelling scientific argument

• The six editions 1859, 1860, 1861, 1866, 1869, 1872

Page 7: Computational Insights and the Theory of Evolution

The Wallace-Darwin papers

Page 8: Computational Insights and the Theory of Evolution

Brilliant argument, and yet many questions left unasked, e.g.:

• How does novelty arise?

• What is the role of sex?

Page 9: Computational Insights and the Theory of Evolution

After Darwin

• A. Weismann

[Paraphrased]

“The mapping from genotype to phenotype is one-way”

Page 10: Computational Insights and the Theory of Evolution

Genetics

• Gregor Mendel [1866]

• Number of citations

between 1866 and 1901:

3

Page 11: Computational Insights and the Theory of Evolution

The crisis in Evolution1900 - 1920

•Mendelians vs. Darwinians

•Geneticists vs. Biometricists/Gradualists

•Population genetics

Page 12: Computational Insights and the Theory of Evolution

The “Modern Synthesis”1920 - 1950

Fisher – Wright - Haldane

Page 13: Computational Insights and the Theory of Evolution

Big questions remaine.g.:

• How does novelty arise?

• What is the role of sex?

Page 14: Computational Insights and the Theory of Evolution

Evolution and Computer Science

• “ How do you find a 3-billion long string in 3 billion years?”

L. G.Valiant

At the Wistar conference (1967), Schutzenberger asked virtually the same question

Page 15: Computational Insights and the Theory of Evolution

Valiant’s Theory of the Evolvable

• Which functions (traits of an organism) can evolve by natural selection?

• Properly formalized, this question leads to identifying obstacles to evolution

• For example, the function has to be learnable (actually, statistically so)

• Evolvability is a (quite restricted) form of learnability

Page 16: Computational Insights and the Theory of Evolution

Evolution and CS Practice:Genetic Algorithms [ca. 1980s]

• To solve an optimization problem…

• …create a population of solutions/genotypes

• …who procreate through sex/genotype recombination…

• …with success proportional to their objective function value

• Eventually, some very good solutions are bound to arise in the soup…

Page 17: Computational Insights and the Theory of Evolution

And in this Corner…Simulated Annealing

• Inspired by asexual reproduction

• Mutations are adopted with probability increasing with fitness/objective differential

• …(and decreasing with time)

Page 18: Computational Insights and the Theory of Evolution

The Mystery of Sex Deepens

• Simulated annealing (asexual reproduction) works fine

• Genetic algorithms (sexual reproduction) don’t work

• In Nature, the opposite happens: Sex is successful and ubiquitous

Page 19: Computational Insights and the Theory of Evolution

?

Page 20: Computational Insights and the Theory of Evolution

A Radical Thought

• What if sex is a mediocre optimizer of fitness (= expectation of offspring)?

• What if sex optimizes something else?

• And what if this something else is its raison d’ être?

Page 21: Computational Insights and the Theory of Evolution

Mixability!

• In a recent paper [LPDF, PNAS 2008] we establish through simulations that:

• Natural selection under asex optimizes fitness

• But under sex it optimizes mixability:

• The ability of alleles (gene variants) to perform well with a broad spectrum of other alleles

Page 22: Computational Insights and the Theory of Evolution

Explaining Mixability

• Fitness landscape of a 2-gene organism

Rows: alleles of gene A

Columns: alleles of gene B

Entries: fitnessof the combination

Page 23: Computational Insights and the Theory of Evolution

Explaining Mixability (cont)

• Asex will select the largest numbers

Page 24: Computational Insights and the Theory of Evolution

Explaining Mixability (cont)

• But sex will select the rows and columns with the largest average

Page 25: Computational Insights and the Theory of Evolution

In Pictures

alleles(variants)of gene A

allelesof gene B

peaks

troughs

“plateau”

Page 26: Computational Insights and the Theory of Evolution

Sex favors plateaus over peaks

Theorem [Livnat, P., Feldman 11] In landscapes of this form

• Unless peak > 2 plateau, in sexual reproduction the plateau will dominate and the peaks will become extinct

• In asexual reproduction, the peaks will always dominate and the plateau will become extinct

Page 27: Computational Insights and the Theory of Evolution

And plateaus accelerate evolution

• They act as springboards allowing alternatives to be explored in parallel…

• …and this acceleration promotes speciation (the creation of new species)…

• …which results in an altered landscape…

• …in which sex selects more plateaus…

• …and life goes on…

Page 28: Computational Insights and the Theory of Evolution

Very Recent [CLPV 2013]Mixability (and more…) established

• In the context of weak selection, evolution becomes a coordination game between genes, where the common utility is precisely mixability (the average fitness of each allele).

• The population stores the mixed strategies…

• The game dynamics is multiplicative updates!

• Besides: Diversity is not lost!

(details in 30 minutes..)

Page 29: Computational Insights and the Theory of Evolution

Pointer Dogs

Page 30: Computational Insights and the Theory of Evolution

Pointer Dogs

C. H. Waddington

Page 31: Computational Insights and the Theory of Evolution

Waddington’s Experiment (1952)

Generation 1

Temp: 20o C

Page 32: Computational Insights and the Theory of Evolution

Waddington’s Experiment (1952)

Generation 2-4

Temp: 40o C~15% changedSelect and breed those

Page 33: Computational Insights and the Theory of Evolution

Waddington’s Experiment (1952)

Generation 5

Temp: 40o C~60% changedSelect and breed those

Page 34: Computational Insights and the Theory of Evolution

Waddington’s Experiment (1952)

Generation 6

Temp: 40o C~63% changedSelect and breed those

Page 35: Computational Insights and the Theory of Evolution

Waddington’s Experiment (1952)

(…)Generation 20

Temp: 40o C~99% changed

Page 36: Computational Insights and the Theory of Evolution

Surprise!

Generation 20

Temp: 20o C~25% stay changed!!

Page 37: Computational Insights and the Theory of Evolution

Genetic Assimilation

• Adaptations to the environment become genetic!

Page 38: Computational Insights and the Theory of Evolution

Is There a Genetic Explanation?

Function f ( x, h ) with these properties:

•Initially, Prob p[0] [f ( x, h = 0)] ≈ 0

•Then Probp[0][f ( x, h = 1)] ≈ 15%

•After breeding Probp[1][f ( x, 1)] ≈ 60%

•Successive breedings, Probp[20][f ( x,1)] ≈ 99%

•Finally, Probp[20][f ( x, h = 0)] ≈ 25%

Page 39: Computational Insights and the Theory of Evolution

A Genetic Explanation

• Suppose that “red head” is this Boolean function of 10 genes and “high temperature”

“red head” = “x1 + x2 + … + x10 + 3h ≥ 10”

• Suppose also that the genes are independent random variables, with pi initially half, say.

Page 40: Computational Insights and the Theory of Evolution

A Genetic Explanation (cont.)

• In the beginning, no fly is red (the probability of being red is 2-n)

• With the help of h = 1, a few become red

• If you select them and breed them, ~60% will be red!

Page 41: Computational Insights and the Theory of Evolution

Why 60%?

Page 42: Computational Insights and the Theory of Evolution

A Genetic Explanation (cont.)

• Eventually, the population will be very biased towards xi = 1 (the pi’s are close to 1)

• And so, a few flies will have all xi = 1 for all i, and they will stay red when h becomes 0

Page 43: Computational Insights and the Theory of Evolution

Any Boolean Function!

• Let B is any Boolean function

• n variables x1 x2 … xn (no h)

• Independent, with probabilities

p = (p1 p2 … pn)

• Now, generate a population of bit vectors, and select the ones that make B(x) = 1

Page 44: Computational Insights and the Theory of Evolution

(cont.)

• In expectation, p p’,

where pi’ = probp (xi = 1 | B(x) = 1)

Conjecture: This solves SATCan prove it for monotone functions

Can almost prove it for weak selection(Joint work with Greg Valiant)

Page 45: Computational Insights and the Theory of Evolution

Interpretation

• If there is any Boolean combination of a modestly large number of alleles that creates an unanticipated trait conferring even a small advantage, then this combination will be discovered and eventually fixed in the population.

• “With sex, all moderate-sized Boolean functions are evolvable.”

Page 46: Computational Insights and the Theory of Evolution

Sooooo…

• The theory of life is deep and fascinating

• The point of view of a computer scientist makes it even more tantalizing

• Mixability helps understand the role of sex

• A natural stochastic process on Boolean functions may help illuminate genetic assimilation and the emergence of novel traits

Page 47: Computational Insights and the Theory of Evolution

Thank You!