Predicting anti-cancer bioactivity of marine sponges using untargeted metabolic profiles

Predictive statistical methods provide powerful but rarely used tools in the search, classification and identification of bioactive compound producing organisms. This in combination with untargeted metabolic profiling can be used to create predictive models that can speed up the process of pre-classification of organisms as bioactive compound producers or non-producers. Sponges are known producers of a wide range of bio-active compounds. One of the many time consuming prerequisites for the identification of sponges that produce bioactives is the activity testing itself. A model that predicts bioactivity can be used to speed up the identification process. To construct a predictive model for bioactivity in marine sponges, samples taken from the South Australian coastline were extracted and analysed by low resolution GC- and LC-MS/ MS to produce comprehensive polar and non-polar fingerprints. In parallel to the fingerprinting, polar and non-polar extracts were tested for anti-cancer activity against eight cancer cell lines (Breast T47D, Cervix HeLa, Colon DLD-1, Prostate DU-145 and PC3, Lung A549, Skin A431, SK-MEL-28). Data matrices containing mass-spectrometry derived chemical information and bioactivity data were used as input for cross-covariance analysis. Various methods were applied to construct predictive models for sponge anti-cancer bioactivity. We illustrate the application of such models to classify untargeted metabolic profiles of South Australian sponges in the search of anticancer marine natural products.