Abstract by Tommy Jensen
Realistic Simulation of Human Population Reproduction
3M Multi-layered perceptron networks have been effective in determining causal factors when enough data is available. These methods could be applied to learning the cause of human genetic diseases, but access to human variants is currently limited to a handful of open projects which provide collections on the order of thousands. This limits the use of multi-layered perceptron networks to learn patterns in unlabeled data. With private SNP chip collections largely inaccessible, the best path to the magnitude of data required is to simulate population reproduction. A robust simulator can produce relatively realistic human SNP collections that bear none of the same legal and privacy burdens a real human population’s records do. New association algorithms can then be developed for use when large genetic databases become available. Our simulator uses provided parameters to force non-Poisson distributed crossover points mimicking real-world measured locations and rates. Variance in recombination rates and inbreeding factors can be successfully verified via linkage disequilibrium measures in our produced populations.