0
$\begingroup$

I have a model that, given a bit code, produces a string result. An example is given 01010101, it might produce {1,11}, and given 01010000, it may produce {2, 11}.

I have a lot of these input / output pairs and I'd like to find, for each output number (the 1, 2, and 11 above), the most likely bits responsible for that number appearing. In the sparse example above, i would expect the algorithm to give as a possible result that 0101____ determines 11.

Note that these pairs aren't necessarily as easy as the example above. For example, there might be multiple bit ranges corresponding to a single number (among different codes).

I don't have experience with genetic algorithms, but this strikes me as being very similar to figuring out what genes are responsible for expressed phenotypes. Is this correct? What is an algorithm that would work for this setting?

$\endgroup$
  • 1
    $\begingroup$ I believe this question is more appropriate for StackOverflow or CrossValidated. You can also ask Mathematica community for help (they are quite helpful) if you want to solve this in Mathematica. Moreover, Genetic algorithms can be used for this but another way to do it would be to use something like SVMs to "learn" the pattern resulting in a phenotype. $\endgroup$ – Siddharth Apr 24 at 12:56
  • 2
    $\begingroup$ I'm voting to close this question as off-topic because it's not about bioinformatics. I agree that different SE might be more appropriate. $\endgroup$ – Kamil S Jaron Apr 24 at 13:22
  • $\begingroup$ Ah I'm sorry. By genetic algorithms I was referring to algorithms for genetics, not Genetic Algorithms. $\endgroup$ – user592419 Apr 24 at 15:20
  • $\begingroup$ Thanks for the feedback. I'll post elsewhere. $\endgroup$ – user592419 Apr 24 at 15:21
  • 1
    $\begingroup$ I'm voting to close this question as off-topic because it belongs on StackOverflow or CrossValidated $\endgroup$ – Scott Gigante Jun 3 at 19:13
0
$\begingroup$

This can be solved using phylogeny quite easily, but it is not clear there would be any biological inference. It might act as a simulation to assess the robustness of various phylogenetic methods. However, "0101____" could create a headache for many phylogeny algorithms and you would need to work alongside someone who knew what the calculation was and its pitfalls.

It wouldn't be something most bioinformaticians would be interested in, because there is a significant risk of zero biological application and essentially would therefore simply become an assessment of the numerical behaviour of an unusual algorithm. Bayesian statistics can have unruly outcomes due to complex reasons - which I wouldn't discuss - and this might be exposed via a simulation.

Genetic algorithms are a computational method, rather than anything to do with biology, which never really took off in genetics although they were briefly implemented.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.