NanoQSAR: metal oxides nanoparticles toxicity assessment.

<p>This is a poster delivered at the MACC-5 – Methods and Applications of Computational Chemistry, Fifth Symposium, 1–5 July 2013, Kharkiv, Ukraine.</p><p>Disclaimers:</p><p>(1) this presentation has not undergone peer review</p><p>(2) this presentation may report preliminary results which may have been revised in subsequent publications</p><p>(3) no endorsement by third parties should be inferred.</p><p>Abstract</p><p>According to the rapid development of nanotechnologies, it is necessary to develop novel methods for reliable predictions of nanoparticles toxicity. Several QSAR studies were conducted recently, but present methods are either rather computationally expensive, or lack the physical understanding of nanoparticles nature. </p><p>Data on acute toxicity towards Daphnia magna and Paranecium multimicronucleatum was collected from different sources. Investigated properties included lethal concentration LC50, mortality and survival ratios, accumulation ratio of metal oxides nanoparticles. In current study the attempt to apply simple physical-based descriptors was made. </p><p>Previously developed descriptors based on liquid-drop model were used, and novel descriptors based on surface-area-difference model were tested. Range of state-of-art modern statistical techniques was applied, i.e. support vector machines, random forest and aggregated neural networks methods. </p><p>Obtained models show good performance R2 for test set above than 0.6, which can be considered quite satisfactory, regarding the quality of experimental data. Developed approaches are shown to be a useful tool to modelling of the toxicity of nanoparticles.