Classification of transformed foci by kernel-based orthogonal projections

Carcinogenesis is a multistep process involving genetic alterations and non-genotoxic mechanisms. The in vitro Cell Transformation Assay (CTA) allows the monitoring of the neoplastic phenotype by foci formation in suitable cells (e.g. C3H10T1/2 mouse embryo fibroblasts) showing typical transformed features. The classification of transformed foci in C3H cells relies on light microscopy scoring by a trained human expert based on standardized rules. Three types of morphologically altered foci have been reported. Cells from type I focus are not considered to be fully transformed. In contrast, cells from type II and III foci are considered to have undergone malignant transformation. The current method of visual classification is widely accepted. Nevertheless this procedure is time- consuming and, in some cases, prone to subjectivity-driven classification. This sometimes may lead to possible over/under-estimation of the carcinogenic potential of tested compounds due to doubtful scoring of foci. Herewith we improved on a recently proposed classification method exploiting “spectrum enhancement” of morphological descriptors by means of Kernel-based orthogonal projection. The ultimate goal of this work is to provide means to classify neoplastic transformation in vitro by integrating image processing and statistical classification. Kernel-based classification is a well known approach widely applied in biological problem domains. The Kernel-based Orthogonal Projections to Latent Structures (K-OPLS) has been recently proposed (Rantalainen et al., 2007; Bylesjö et al., 2008) to offer both nice predictive performances and enhanced interpretational capabilities. Systematic variation like instrumental drift and batch variability are captured into orthogonal components while non-linearity of features is accounted by exploiting suitable kernels.