AG Kommunikationstheorie


Application of Information-Theoretic Measures in Genetics


Since the discovery of the deoxyribonucleic acid (DNA) as carrier of the genetic information, and the decipherment of the coding of the proteins, there has always been great interest in connections between the genome and the phenotype, the outward appearance of an individual. Several results have since then been achieved in gene mapping, the science of finding these dependencies. Of course the dependencies that were the easiest to find are the so called Mendelian traits, where the phenotype trait depends on a single genetic marker in a deterministic manner.

But as a lot of mechanisms in gene expression, the transformation of genes to proteins, are not yet understood, it is no surprise that a lot of dependencies are yet to be discovered. This are first of all more complicated dependencies, where multiple genetic markers are involved, maybe along with some random influence. In recent years, methods from information theory have proven to be valuable tools for finding these dependencies, which are very hard to discover otherwise.

An important problem in the application of information-theoretic measures is the estimation of these measures (for example mutual information) from samples, usually without knowledge of the underlying distribution. Furthermore, the distribution of the estimator itself is essential for the evaluation of the results, for example in the context of hypothesis testing.

In this talk, some established approaches for the estimation of the mutual information will be presented, along with methods to approximate the distribution of the estimator. Comparison with simulation results will show how well these methods perform. Furthermore, the application of a different approach using spline functions will be examined by simulation.

zurück zur Terminübersicht