Neural networks for computing invariant clustering of a large open set of DNA-PCR-primers generated by a feature-knowledge based system

A neural network computing model for image sorting by means of invariant clustering is described. Such image sorting by clustering belongs to the class of inverse problems of image processing. One of the biomedical applications is to diagnose AIDS virus-mutated DNA by a powerful recombinant DNA technology, the polymerase chain reaction (PCR), based on judicious choice of oligonucleotide primers for the synthesis of an imprecisely known DNA sequence due to the unknown viral infections. Thus, a smaller set of primers among a larger open set, generated by a feature rule-based system, must be found. A neural network is needed to do nondestructive sorting and maintain the chemical bases (G, C, A, T) of the chosen primers. The problem is solved by an ART-1 neural network for feature extraction by clustering and an ART-4 (a hybrid ART-1) neural network for image sorting by clustering. The initial condition of ART-1 is precisely the complement of the initial condition of ART-4. The ART-1 whitens the cluster leader by the veto power of newcomers, while the ART-4 edifies and preserves the cluster leader by the consent of followers. A feature based expert system provides a powerful training environment for the ART-4 neural network to further discover new cluster rules