Improved and Novel Cluster Analysis for Bioinformatics, Computational Biology and All Other Data

摘要

Cluster analysis has been widely used in bio-informatics or biology to classify objects (items, DNA bands, markers, genes, individuals, species, taxa, etc.). Theoretically, there are numerous clustering methods available but only those that are well-established and proven are commonly used in practice. For their improved applicability, those methods that exploit the most data information and yield the better cluster properties should be focused and new algorithms are expected. The other issues and problems that are relevant to the performance of cluster analysis also need to be addressed. The well-prepared data can be significant to ameliorate the clustering results. They include the proper measure conversion from similarities to dissimilarities or distances and the necessary data standardizing transformation. Apart from z-score standardization, average- or mean-based scaling was introduced to achieve comparability among variables with the least value manipulation. A solution was given to improve updating the distance matrix with a centroid-related equation. A novel method called the percent similarity cluster analysis (PSCA) was devised to analyze the DNA band or marker data from electrophoresis and all other data.