This is a proposal to extend substantially a method of structural pattern analysis of human G-banded chromosomes by computer. A database of 6985 digitized profiles of chromosomes will be used. Individual profiles are filtered and mapped into a string representation which emphasizes band transitions. Dynamic programming is used to compute optimal (maximum probability) string alignments, both to infer a model (a finite Markcv chain) from learning samples and to classify by chromosome type in recognition experiments. Use of this method will be extended to more detailed analysis of structure. In particular, automatically determining p-q vs. q-p orientation and locating the centromere will be investigated. To increase the speed of computation, constraining the dynamic programming such that unlikely alignments are not computed and reducing the size of networks by removing nodes labelled with very low frequency of occurrence will be investigated. To make use of higher level knowledge about the chromosome complement of human cells, the structural models will be incorporated into a knowledge-based program which uses the structural analysis together with the knowledge base for analysis at the level of individual cells. This will be a step towards an integrated system of multilevel analysis. This work will further the understanding of structural pattern analysis of the band patterns and will increase the information about individual chromosomes provided by this analysis. Experiments will use on the order of 200 samples of each of the autosomes.
|Granum, E; Thomason, M G (1990) Automatically inferred Markov network models for classification of chromosomal band pattern structures. Cytometry 11:26-39|