You are on page 1of 31

Avida: Increasingly complex functions

adapting to novel basic environments


Emily B., Allison A., Graham B., Jordana S.
BIOL 481
5 December 2019
Why use Avida?
● 2000 generations, or updates, take about 5-10 minutes.
● Genotypes, in the form of function capability, can be easily manipulated.
● It is free!
● There is no waste or live organisms associated with experiments.
● 9 possible functions of various difficulty can be rewarded independently or in
combination with each other to create unique selection pressures.
How do the adaptation rates of complex populations
of Avidians compare to simpler populations in a
novel basic test environment?
HISTORICAL CONTINGENCY: Will the evolutionary background of a population
matter?

If it does...

Will more complex populations have a higher adaptation rate in a novel basic
environment?

Or will simpler, less evolved populations have a higher rate of fitness increase?
If evolutionary background does not matter...
NO SIGNIFICANT COSTS: There is no significant difference between the fitness
cost of simple and hard functions.

COMPENSATORY MUTATIONS: Complex populations may have acquired


compensatory mutations which offset fitness costs associated with hard functions.
If more evolved, complex populations have a higher
adaptation rate in a novel basic environment...
SURVIVAL OF THE MODERATES: Selection pressure against complex Avidians
may be stronger than the selection pressure against moderate Avidians, leading to
a higher rate of extinction of complex individuals from the population.

COMPLEX GENERALISTS >> SIMPLE SPECIALISTS: Avidians with complex


functions are inherently able to perform simple ones as well, and are thus more
suited to a novel basic environment than a population evolved in a specific basic
environment that does not match.
If simpler, less evolved populations have a higher
adaptation rate in a novel basic environment...
Complex functions are costly to maintain when they are not rewarded.

SIMPLE GENERALISTS >> COMPLEX SPECIALISTS: Populations with fewer


complex functions can acquire simple functions more easily because they do not
have to expend resources on complex functions.

STREAMLINED GENOME: Highly evolved populations may become more


efficient than moderately evolved populations, and experience less negative
selection pressure than simpler populations. Thus, complex individuals would
remain in the population longer than moderate individuals.
Experimental Overview
Update 2000 Update 4000
Ancestor 1 2
Avidian NAN Moderate Moderate Hard
or Functions Functions Functions
NOT

Update 6000 500 updates


3 Very Hard
Hard NAN or NOT (opposite
Functions Functions
starting) test environment
Experimental Overview - Avida Functions

1 - NOT (easy) 5 - ORO (hard)

2 - NAN (easy) 6 - ANT (hard)

3 - AND (moderate) 7 - NOR (very hard)

4 - ORN ( moderate) 8 - XOR (very hard)


Experimental Overview - Evolution Phase 1
Initial setup:
● 30x30 dish size
● 2% mutation rate
● Ancestor Avidian
● Random generation location
● One basic function + both moderate functions rewarded
● 2000 Updates

This setup at 2000 updates denotes population 1_3_4 or 2_3_4


depending on which basic function (NOT or NAN) begins the trial.
Experimental Overview - Evolution Phase 2
Second setup:
● 30x30 dish size
● 2% mutation rate
● 1_3_4 or 2_3_4
● Random generation location
● Both moderate functions and hard functions rewarded
● +2000 Updates (total 4000)

This setup at 4000 updates denotes population 3_4_5_6.


Experimental Overview - Evolution Phase 3
Third setup:
● 30x30 dish size
● 2% mutation rate
● 3_4_5_6
● Random generation location
● Both hard functions and very hard functions rewarded
● +2000 Updates (total 6000)

This setup at 6000 updates denotes population 5_6_7_8.


Experimental Overview - Evolution Phase - What did
we just do? What’s next?

● Elevated number of complex functions


● Decreased number of simple function
● Without the simple functions, complex functions take too
long to evolve!
● So what do we do with these populations? Compare them
in a constant, novel, basic environment!
Experimental Overview - Test Phase

Test Environment 1 Test Environment 2


2_3_4 3_4_5_6 5_6_7_8 1_3_4 3_4_5_6 5_6_7_8

500 updates run for each population 500 updates run for each population
separately in novel basic environment separately in novel basic environment
(here test environment = NOT, function (here test environment = NAN, function
1) 2)
Experimental Overview - Controls
Does the presence of a novel basic function in the test environment alter
the initial fitness compared to a non-selective environment?
- All frozen down populations enter a non-selective environment for one
update to measure initial average fitness.

What if you have no functions when placed in the test environments?


- Ancestor is placed in both the NAN and NOT test environments for 500
updates.
moderate

simple complex

basic, novel, environment


moderate

Predictions
simple complex

basic, novel, environment

H0 H1 H2
Complexity → Complexity → Complexity →
who cares? generalization specialization
Data Analysis
● We ran ANOVA, t-tests on 3 metrics we had collected.
○ Initial fitness in test environment
○ Final fitness in test environment
○ Change in fitness in test environment
○ We ran this both individually on the NOT groups and the NAN groups, as well as combined
them
● What is ANOVA?
○ “Analysis of Variance”
○ Can be thought of as similar to a t-test, but among more than 2 groups
“Without control” results
● Without a control and when NAN/NOT groups aren’t combined, differences
between groups (i.e. 1_3_4, 3_4_5_6, etc.) were not statistically significant. (p
> .05)

1, 3, 4 + 3, 4, 5, 6 + 5, 6, 7, 8 = no difference in fitness performance

2, 3, 4 + 3, 4, 5, 6 + 5, 6, 7, 8 = no difference in fitness performance


With a control
+1, 3, 4 + control = significant fitness difference

P << .001 in every case,


2, 3, 4 + control = significant fitness difference for Δ fitness, initial
fitness, and final fitness

3, 4, 5, 6 + control = significant fitness difference

.
.
.
When NOT and NAN groups are combined
● Now, we have only 3 distinct groups: one made of X_3_4, another of
3_4_5_6, and 5_6_7_8.
● Statistically significant results found for initial fitness, nothing significant found
for final fitness and Δ fitness
● T-tests performed on the initial fitness results. Difference between X_3_4 &
3_4_5_6, and X_3_4 and 5_6_7_8 are statistically significant (p < .01).
**
****
Interpretation of results
● Initial fitnesses increase up until “medium” complexity, but medium & hard
have no difference. Avidians don’t seem to be hindered by evolving in
complex environments, so far. This might be because
○ Basic functions are required steps that get built in?
○ Basic functions aren’t particularly helpful in making complex ones, but it’s hard to lose a
function that doesn’t decrease fitness at all/significantly within only 2000 steps?
○ Medium/hard populations have simply evolved for longer than easy?
● NO difference in delta fitness, final fitness: if a population starts off with even
one individual that has NOT/NAN, then the function can fix quickly.
Alternatively, all functions are equally accessible--but this is probably unlikely
● Control population performs worse than treatment populations
Interpretation, continued
● Is it easier/harder to access NAN/NOT in more “complex” populations?
○ Hard to say, because we transferred entire populations, and many of them may have had
preexisting NOT/NAN individuals in the population, and regardless, we found no difference in
Δ fitness. We only found a difference in initial fitness, which has nothing to do with
accessibility
● What we do know is that even when certain basic functions are not rewarded,
populations exhibit a higher starting fitness in environments that reward those
basic functions. However, the difference disappears once enough time has
passed.
More Science!
● More entrenchment ● Less overlap

● Less diversity ● More capacity

● Better analytics

You might also like