Tools and applications of Genetic Linkage Analysis (NNN.6414)

One of the major questions in genetics is to find the genetic origin of diseases in humans and animals and to understand how genetic mutations may alter particular properties in animals and plants. A specific example is genetic linkage analysis where one attempts to discover correlations between the occurrence of a disease in family pedigrees and the presence of one or more mutated genes on the chromosomes. These studies require models that specify the mechanisms of inheritance and the effect that mutated genes may have on the disease. Such a model assumes the disease to be present at a certain locus on the chromosome. The locus that belongs to the model that best explains the observed experimental data (marker data and affection status of the individuals) is the most likely position to find a genetic defect.

Due to the inherent statistical nature of the problem and due to the large number of variable involved, the method of choice is the use of probabilistic networks. The important advantage of this approach over existing methods is the transparency of the model assumptions, which are often obscure in alternative approaches. In addition, it is clear how to incorporate multi-locus diseases (epistasis) in the probabilistic approach or to incorporate additional (prior) knowledge, such as population statistics, etc. which is extremely difficult in the existing approaches.

An important problem for all techniques (including probabilistic networks) is the computational complexity. Even for rather small families or a small number of markers it is intractable to do exact computations. As a result of improved experimental techniques, the amount of available genetic data and variables that require modelling vastly increases, making the computational problem still more urgent in the future.

By treating genetic linkage analysis as a Bayesian inference problem we can apply existing techniques from the field of machine learning. One can distinguish methods that 1) improve the speed of exact calculations, 2) yield approximate results end 3) upper and lower bound the value of quantities of interest. We thing that each of these techniques has a promising application to this problem.

The methods will be developed in the context of a number of concrete applications. With Nutreco, we will explore applications in the area of animal breeding programs. With our methods we aikm to gain insight about the genetic origins of animal susceptibility to disease, which can lead to more effective breeding programs and healthier animals.

With Keygene, we will explore applications in the area of food design and food production. The aim of this research is to gain more insight in the genetic mechanisms that affect plant resistance to diseases as well as their nutritional value. which may have important implications for future food products.

With decode, we will re-analyze some of their large data sets that this company has collected from the human population of Iceland.
Clearly, the results of this research will be important for many other companies and research institutes, as well as for other application areas (for instance design of new drugs). We will develop computer software that will allow other parties to use the methods developed in the project.