Artificial beings evolve realistically

By Kimberly Patch, Technology Research News

For the past couple hundred years, scientists have studied genetics using organisms like peas, fruit flies and viruses, whose lifespans are much shorter than ours.

In the past decade or so, computer scientists have worked with genetic algorithms, which can produce many generations much more quickly and can track each change more closely than is possible using real organisms. But these virtual organisms evolve much more simply than the real thing.

Researchers from Michigan State University, the California Institute of Technology and the University of California at Los Angeles have found a way to use software to more closely mimic the way real organisms evolve, and have used the model to uncover a long-standing secret of natural selection. "Our goal is to better understand... how genes form," said Charles Ofria, an assistant professor of computer science at Michigan State University

A better understanding of how DNA evolves is useful on many fronts: DNA's methods could make software more stable and communications lines more efficient. The information promises to help medical researchers design more efficient drugs and better track the ways environmental stresses like radiation affect genes. And a better understanding of evolution at the DNA level could help biologists more accurately reconstruct evolutionary relationships.

The researchers tapped information theory to uncover the balance of evolutionary pressures that determine which genes are transmitted from one generation to the next. "We conceptually treated a living organism as an information channel and the information it transmits as its genome from itself to its offspring," said Ofria.

This allowed them to isolate the evolutionary pressures that involve the transmission of genes from the evolutionary pressures that affect whole organisms.

The researchers' digital organisms are simple computer programs that self-replicate. "You can think of them as computer viruses that are well-contained within the computer," said Ofria.

The experiments isolated three distinct pressures that produce replication advantages: compression, transmission and neutrality. The first two pressures involve the number of genes passed on to the next generation.

The lower the number of genes, the faster and less resource-intensive it is to transmit those genes. This gives an advantage to organisms with fewer genes, which, in turn, tends to limit the average number of genes in a population. This is compression pressure.

On the other hand, including as much useful information as possible about the environment in the genome conveys a different advantage, said Ofria. "If an organism can interact well with its environment -- avoid predators, go and find food that it needs and things like that -- it will live longer and produce more offspring," he said. This is transmission pressure.

The first two pressures balance each other. "If there's more information in the genome, the genome has to get longer," said Ofria. "In more complex organisms this is the pressure that ends up dominating. In very simple organisms like viruses, it's [compression] pressure that dominates," he said.

The third pressure, neutrality, keys off the amount of redundant information a genome contains a. Redundancy allows organisms to better withstand mutations.

Genes are sequences of the four bases that make up DNA, and are blueprints for proteins, which do much of the work of maintaining life. If a genome has a lot of room for mutation, meaning if a base changes here and there and the gene will still code for the correct protein, it has a high fault tolerance, or neutrality.

Neutrality is built into living organisms in the way genes are encoded. Three bases provide 27 combinations, which is enough to code for the 20 amino acids that form the building blocks of proteins. Instead, however, organisms use strings of four bases, with 64 combinations, which makes the code more resistant to mutation.

Genomes also contain higher-level coding characteristics that are less well understood. Sometimes when one portion of a gene gets changed, a completely different section will compensate.

The researchers' digital organisms contained about 100 lines of code and lived in simple environments where they were able to self-replicate. Each experiment contained 3,600 organisms, and ran for 10,000 generations.

In the simplest experiment, the organisms simply had to replicate. There were no environmental differences, and thus no transmission pressures, and the neutrality pressure was very weak. The 100-line organisms quickly shrank to around 21 lines of very efficient code, said Ofria. "As they make themselves shorter there's less for them to copy, [and] the more offspring they can produce per unit of time," he said.

In a second set of experiments, the researchers used the 21-line organisms and introduced transmission pressure using a reward system that allowed an organism to get ahead by performing calculations like addition. "If they take two numbers and they add them together and output the results, we would give them a little bit more CPU time. If the amount of additional CPU time was greater than the time it took to add two numbers together, then the amount of time left over they can put toward self-replication," said Ofria.

The key was rewarding the organisms not for how they performed the task, but for simply taking the numbers in and outputting the correct result, said Ofria. "In that sense the evolution is open-ended."

At the same time, the researchers removed compression pressure by giving the organisms additional CPU time in proportion to their length. This balanced the advantage of becoming smaller, he said. There was still some pressure not to be too long, because longer organisms were more susceptible to mutation.

Three transmission-pressure environments -- no calculation, a moderately difficult calculation, and a complex calculation -- bore out expectations.

In the simple environment the organisms grew only a little, to 23 lines, and the replication rate stayed the same. In the medium environment organisms replicated more quickly and increased to 54 lines. In the complicated environment they grew to just under 100 lines and replicated much more quickly.

The third experiment, which used the final organisms from the complicated environment and introduced mutations into the mix, began to tease apart a long-standing chicken-and-egg problem concerning neutrality.

Mutations, or random code changes, are a double-edged sword. They are often detrimental, but are sometimes do no harm, and can speed the process of it after experiments show evolution.

There are competing theories about which edge of the sword is the more important source of the pressure for organisms to increase fault tolerance, or neutrality.

In the third set of experiments, a low mutation rate decreased neutrality, but a high mutation rate nearly doubled neutrality.

The researchers are tracing back the lines of descent to see exactly which changes increase neutrality, according to Ofria. The program allows the researchers to look at every organism's entire ancestral line, he said. "We can go to each child-parent pair... see what the difference is between the child and parent... and see what changed in the underlying program and how that change [made] things more neutral."

The researchers' preliminary results showed that fewer detrimental mutations are better in the short-term, but a larger number of neutral mutations have more long-term advantages, Ofria said.

In general, being able to trace exact lines of dissent enables a much closer look at gene interaction than is possible using natural systems, Ofria said. "Working with something like bacteria you can get a lot of generations, but you can't possibly keep track of it with much detail. And working with any larger animals the generations just take so long that you could never have evolved so much," he said.

In general, the method should help researchers better understand what's going on inside the genome and thus what to look for in real genes to decipher how they interact, Ofria said.

"What we want to do is not so much directly understand the genetic code, because that's almost impossible," said Ofria. But understanding evolutionary pressures gives researchers a "better idea of what we would expect to be going on," he said. "We might see certain patterns of gene regulation that we can go look for in natural systems."

The digital organisms might also provide some lessons about robust computer code, said Ofria. "I'm hoping to generate some principles for robust coding... that programmers would apply," he said.

The method could also improve genetic algorithms, which are modeled after evolution. The algorithms do not really employ natural selection, however, because the organisms do not have to self-replicate, and therefore are not affected by the transmission pressures, said Ofria. The organisms whose traits are passed on are automatically chosen based on their traits, generally because those sets of traits contain the best adaptations to environmental conditions.

One problem with genetic algorithms is that they can get too complicated before they reach a solution. The addition of natural selection weeds out organisms that are too fragile to self-replicate, which reduces the complexity.

The method is "quite good," said Charles Taylor, a biology professor at the University of California at Los Angeles. "The notion of information being used in the genetic code and subject to evolution has been appealing for many years [but this research] group is the first I have seen," that has tapped the idea in a useful way, he said.

The researchers' method is potentially useful for population genetics and for analyzing phylogeny, he said. "Both allow questions of the sort 'how much do I know about "x" based on other information I might have,'" said Taylor.

Another research team has used the simulation to examine exactly how complex functions evolve incrementally from simple changes. (See "Simulated evolution gets complex," TRN May 21/28, 2003)

The researchers' next step is to use the method for a practical biological purpose. "We're looking at problems of phyllo genetic tree reconstruction, which is like trying to reconstruct the tree of life," said Ofria.

The method could be used in practical computer science applications within two years, Ofria said. Practical biological applications are 10 to 15 years away, he said.

Ofria's research colleagues were Christoph Adami from the California Institute of Technology and the Jet Propulsion Laboratory and Travis C. Collier from the University of California at Los Angeles. The research was funded by the National Science Foundation (NSF).

Timeline:   2 years, 10-15 years
Funding:   Government
TRN Categories:  Applied Technology; Artificial Life and Evolutionary Computing
Story Type:   News
Related Elements:  Technical paper, "Selective Pressures on Genomes in Molecular Evolution," posted on arXiv physics archive at


June 4/11, 2003

Page One

Shock waves tune light

Artful displays track data

Plastic transistors go vertical

Artificial beings evolve realistically

News briefs:
Microfluidics go nonlinear
Browser boosts brain interface
Semiconductor emits telecom light
3D display widens view
Study shows DNA will fill tubes
DNA part makes transistor


Research News Roundup
Research Watch blog

View from the High Ground Q&A
How It Works

RSS Feeds:
News  | Blog  | Books 

Ad links:
Buy an ad link


Ad links: Clear History

Buy an ad link

Home     Archive     Resources    Feeds     Offline Publications     Glossary
TRN Finder     Research Dir.    Events Dir.      Researchers     Bookshelf
   Contribute      Under Development     T-shirts etc.     Classifieds
Forum    Comments    Feedback     About TRN

© Copyright Technology Research News, LLC 2000-2006. All rights reserved.