Detecting Similarities between Families of Bio-sequences using the Steady-State of a PCA-Neural Network

In this paper we proposed a novel algorithm to detect similarities and dissimilarities between any two given families of biological sequences using the steady state concept of a PCA-neural network. Traditionally, PCA is an encoding technique used to encode data embedded in a high dimensional space to a lower dimensional space under the condition that a minimum loss of information must be achieved. Training a PCA-neural network with two stationary multivariate stochastic processes sequentially may lead the PCA-neural network to converge to the same attractor point or to two different attractor points, and thus this valuable equilibrium property of the PCA neural network can be employed to solve the similarity detection problem empirically. The performance of the proposed algorithm shows robustness and accuracy in similarity/dissimilarity detection