Last week we had a chance to try some password puzzles. All of them involved different variations of exploration and feedback. Some of the variations made the challenge harder than others; hopefully if you had a chance to play that became apparent. I wanted to help build some intuition for how widely the difficulty can vary. We can also think of these as possible models for the challenge posed to evolution at the molecular level. So the question is which if any of these models fits best, and do any of them fit well?
The basic analogy linking molecular evolution to all those models is the fact that genes and proteins have a linear sequence. We discussed that two weeks ago for genes and genomes; the DNA molecules which represent them are long chains of four varieties of nucleotides. Working out the series of nucleotides that make up a gene thus has some similarity to working out the sequence of characters in a password.
At the same time, the way that biological systems can tackle that challenge differ from all the versions of our puzzle in a few ways. You had the chance to see everything you had previously tried and thus avoid repeating possible solutions, but organisms generally have no mechanism for storing alternate versions of their genome. You could choose what solution to try next based on what had worked or not worked previously, but mutations in organisms generally follow the same patterns regardless of context or history. And you could stop trying when you were done, whereas functional genes always have the chance to mutate and there is no concept of done because which what makes a sequence functional is dependent on context that can change. Still, with the understanding that our models all have limits, we can still explore how they relate to biological evolution.
The easiest version of the challenge was the one with the strongest feedback, where individual correct letters were called out and could not be changed subsequently. That one was too easy compared to biological scenarios. Crucially, that puzzle could be solved in at most 26 tries (just type all a’s, then all b’s and so on until all the letters are filled in) regardless of length, and we know empirically that genes cannot evolve in quite that way. Still, depending on the context, negative selection can be quite strong. So while an individual organism might have a mutation that changes a nucleotide which is crucial to function, with strong negative selection that mutation will be removed from a population and not passed on for generations and generations. In the long run, the effect is similar to ‘freezing’ individual letters.
Likewise, the hardest version is harder than at least some biological scenarios. Biochemical function is not an all-or-nothing proposition. There can be partial ‘credit’ for an inefficient solution, which can be further optimized via additional rounds of evolution. Consider for example the challenge of generating antibodies which we’ve discussed before. In order for the process to start, the set of antibodies we were born with has to include at least one that binds partially to the infecting pathogen. From there, the binding can be refined until it is quite strong, but if there were no binding at all at the start then no B cell would get the signal to proliferate. When gene activity has a range of efficacy like that, the intermediate feedback scenario is the closest model.
Finally, we can ask the question of whether there are better models. Over the last few posts, I have deliberately avoided writing DNA sequences using the letter abbreviation convention of A for adenine, C for cytosine, G for guanine and T for thymine, so a sequence looks like AGACAGATAGGATTACATACGATGA. It’s a reasonable convention, much more efficient than using the full name every time. But it does also lead us down the garden path of making analogies between nucleotides and letters, which then leads to thinking of evolution as a spelling exercise or a password cracking scenario. And it’s a short hop from there to monkeys and typewriters, a model equivalent to the minimal feedback scenario from last week. We also get examples like Richard Dawkins’ famous WEASEL program, which does illustrate selection but also reinforces this idea that evolution is about looking for that one correct spelling. What if we stayed with language but tried some other tasks besides spelling prespecified words? What other language puzzles could we evolve a solution to? Give that some thought, and next week we’ll try to explore some alternatives. If you have an idea, feel free to leave it in the comments and I’ll see if I can program it for us to play with.