Editorial for the special issue of knowledge discovery and management in biomedical information systems

The rapid growth of research and development using different advanced techniques, e.g. AI and Machine Learning, in biological and medical information systems has drawn worldwide attention on the management of knowledge discovered in, for instance, gene databases and public healthcare portals. However, challenges remain from various fundamental issues, e.g. difficulty in data acquisition and heterogeneity of biological information, to the lack of methodology that supports the fusion, processing and management of abundant biomedical and genomic data and eventually the management and interpretation of knowledge that arises from such discoveries for the purpose of problem solving and decision making in translational medicine, e.g. clinical medical practice. This Special Issue is therefore dedicated to innovative, state-of-the-art research, technology development and applications of knowledge discovery and management in biomedical information systems. All the papers included in this issue can be basically clustered into two groups. The first group mainly focuses on using data mining in assisting disease prediction, medical diagnosis, and rule discovery in microarray data, etc. “Efficient Mining of Multilevel Gene Association Rules from Microarray and Gene Ontology” by Tseng, Yu and Yang, presents an algorithm that extended the current association rule mining with the assistance of the concept hierarchy in gene ontology to discover multilevel gene association rules from microarray data. In particular, a hierarchy information encoder has been utilized to provide the structured vocabularies which are organized in a rooted directed acyclic graph interpreting the roles of genes and gene products. This encoder serves as an interpreter of domain knowledge and it helps to limit the number of rules generated. Their experimental study shows that the proposed method is able to provide biologists different rule sets which indicate different gene expression patterns existed in cells. The refined version of the proposed algorithm can further interpret the prior biology knowledge into constraints embedded in the rule mining. “Comparing Data Mining Methods with Logistic Regression in Childhood Obesity Prediction” by Zhang, Tjortjis, Zeng, Qiao, Buchan and Keane, reports a study using non-linear interactions to help improve prediction accuracy of children obesity by comparing the results of logistic regression, a prevailing choice in medical prediction analysis, with some six well established data mining algorithms which are commonly used for either prediction or classification purpose, e.g. decision tree and decision rule, Bayesian network and SVM. The empirical study based on real-world data demonstrates that SVMs can Inf Syst Front (2009) 11:345–347 DOI 10.1007/s10796-009-9153-4