CSAIL Event Calendar

Evolution Analyzed as a Communication Channel: Capacities and Codes

This talk presents an a priori information-theoretic analysis of evolution, by defining and calculating channel capacities of some standard genetic models. This work is relevant to genetics and evolutionary computation.

Organisms are wonderfully adapted to their environments. Some natural basic questions are: How complex can evolved organisms become - what sets the limit? What precision of adaptation is possible? Are there different limits on the precision of adaptation of sexual and of asexual organisms? What are the most efficient ways to encode information in the genome, so that the greatest amount of information can be maintained with the smallest amount of selection? Are efficient genetic codes different for sexual and asexual organisms? How, indeed, can we define the 'complexity of an organism'?

This talk (partially) answers these questions by setting out a clear framework for analyzing selective breeding as a communication channel: the capacity of this channel is meaningful as a limit on the achievable precision of adaptation. The capacities of some standard genetic models are calculated or estimated, using this framework. This framework sidesteps the ill-posed question of how to define the 'complexity of an organism' by considering instead the channel capacity of a selective breeding experiment, which is a well-defined quantity that can actually be calculated.

A surprise is that, for sexual reproduction, the greatest channel capacity is achieved for long genomes in which individual loci are poorly controlled by selection. This suggests that, for sexual organisms, the 'junk' is the part of the genome in which information can be most efficiently accumulated and maintained by evolution. For asexual organisms, there is no advantage in such diffuse genetic encodings.
The reason that 'junk' may be informationally efficient for sexual organisms is formally similar to the reason that, in radio transmission, wide-band encoding is more efficient than narrow-band for a power-limited Gaussian channel.

The performance of some error-correcting codes in standard genetic models is presented: some well-known error-correcting codes can allow greater amounts of information to be maintained in the genome with less selection than is needed without error correction. This raises the question of what other efficient genetic codes are possible, and how they might evolve.

See other events happening in November 2009