Abstract
Current matching procedures to select donors for transplantation are time consuming and expensive. We have developed a novel DNA based method of tissue typing termed MHC block matching. The technique not only matches the class I and II genes of the human Major Histocompatibility Complex (MHC) but also the alleles at yet undefined loci that affect transplantation outcome. The densitometric profile (obtained after PCR and scanning with a laser on a Corbett Research GS 2000 Gel Scanner) from donors are compared to that of potential recipients. Differences in the profiles of the donor and recipient are biologically relavent due to sequence differences in the MHC genes and they correlate to outcome. Our aim is to identify critical regions of the profile which have to be matched for successful transplant outcome. Thus, in an unrelated context, we are able to recognise permissible mismatches as well as important genes for transplantation. This presentation describes the use of a decision-trees, a feature-based machine learning technique, that learns from example the biological relavent regions in profile pairs that contribute to outcome. We will discuss the method of feature extraction to obtain a set of characteristic features for a given set of profile pairs and report on our initial results from constructing a decision tree and dealing with issues of noise in the training set.