Biological systems are extremely complex and often involve thousands of interacting components. Despite all efforts, many complex biological systems are still poorly understood. However, over the past few years high-throughput technologies have generated large amounts of biological data, now requiring advanced bioinformatic algorithms for interpretation into valuable biological information. Due to these high-throughput technologies, the study of biological systems has evolved from focusing on single components (e.g. genes) to encompassing large sets of components (e.g. all genes in an entire genome), with the aim to elucidate their interdependences in various biological processes. In addition, there is also an increasing need for integrative analysis, where knowledge about the biological system is derived by data fusion, using heterogeneous data sets as input. We here review representative examples of bioinformatic methods for fusion-oriented interpretation of multiple heterogeneous biological data, and propose a classification into three categories of tasks that they address: data extraction, data integration and data fusion. The aim of this classification is to facilitate the exchange of methods between systems biology and other information fusion application areas.A key issue in DF of heterogeneous biological data is that there is no gain from fusion if no increase in understanding of the underlying biological system is achieved. As discussed in [7], the results of IF should provide effective support for humans in their work to make decisions. Therefore, prior to DF it is critical to extract biologically meaningful features and detect the biologically relevant correlations between these features [19]. When inferring biological networks we work both with methods for assigning functional annotation to each of the elements in the network and with methods for analyzing causal effects between those elements. These tasks require use of data from various data sources by means of appropriate methods [2]. Here we report on selected methods for network inference [1,2,5], pathway reconstruction [19,20], binding prediction [21] and functional annotation [22], all of which can be considered as examples of methods from the DF category. To certify that we have a representative selection of DF methods we have chosen methods that use different approaches such as correlation calculation, Bayesian network, clustering, multivariate regression, machine learning, combinatorial analysis, and statistical tests in the DF process.In a study by Yamanishi et al. [19] a novel method was proposed for detection of correlation between three different biological data sources (functional data, geometrical data and co-expression data) with the aim of reconstructing pathways. The authors also emphasized the importance of first detecting biologically relevant correlations before fusion of data. The proposed approach consists of a generalized kernel canonical correlation analysis (KCCA) and a method for extraction of groups of genes responsible f...