• Poster presentation
• Open Access

• Published: 05 June 2009

Conventional QSAR/QSPR models are built only for one target property without exploiting any a priori knowledge stored in datasets of related properties. Here, individual models are not viewed as separate entities but as nodes in the network of interrelated models. Such interrelated models can be built in parallel by means of multitask learning (MTL), or sequentially using feature nets (FN). MTL and FN are kinds of data integration, as opposed to traditional single-task learning (STL), in which all models are built separately. We apply this strategy to model Human blood:air, human and rat tissue:air partition coefficients of organic compounds using diverse and relatively small datasets.