Guest commentary (1) on chapter 10: Towards building knowledge-based assistants for intelligent data analysis in biomarker discovery
The key challenges and potential research directions for bioinformatics and disease biomarker discovery highlighted in Chapter 10 of this book well describe the most important issues that the field should face in the next few years. Amongst such challenges, a very intriguing one for bioinformaticians is related to data analysis methods to support biomarker discovery. As properly reported by Francisco Azuaje, the extraction of multivariate predictive models based on machine learning and statistical techniques may effectively provide tools for diagnosis and prognosis based on apanel of biomarkers, or ‘omic’ signatures. Since the end of the last century several approaches based on gene expression microarrays (Brown et al., 2000) and on mass spectrometry data have been proposed (Yu et al., 2005). However, several of those approaches suffered from lack of reproducibility; moreover different models with the same prediction capability starting from the same data set were derived. As a matter of fact, this problem is related to both experimental and data analysis pitfalls (Hu, Loo and Wong, 2006). As far as the data analysis problems are concerned, the automated extraction of multivariate models is heavily constrained by the intrinsic limitations ...
