The completed progress in regards to the Phase I of the project entitled "Molecular-level prediction and mitigation of side effects of cancer therapy drugs" includes the development of a high-quality benchmark dataset, and first-of-its-kind evaluation of state-of-the-art inverse docking predictors, Findsite and SMAP, on this dataset. Our results show that these two methods perform relatively well, however further improvement are needed to allow for accurate ADR predictions. Importantly, we also developed a novel consensus predictor that combines Findsite and SMAP, which outperforms both of these methods. These preliminary results are very encouraging and will be submitted for high-impact journal publication in 2011. .However, our analysis shows that the current solutions, including our consensus solution, are not accurate enough for the prediction of adverse drug reactions (ADRs). Moreover, the results on our dataset reveal a novel way to build a new generation of the inverse docking predictors. Specifically, we observe finding similarity in the binding pocket would allow for improved prediction. This similarity can be described using recently proposed Zernike moments (Kihara et al., 2011; Saeland Kihara, 2010). This approach would complement the currently utilized (by Findsite) fold similarity and (by SMAP) sequence similarity. As our goal 1 for the Phase II, we plan to develop a new, highly-accurate inverse docking predictor that combines the complimentary predictions from Findsite, SMAP and our pocket similarity-based solution. Our preliminary results from Phase I show that such system could generate predictions of sufficient quality that would allow for application into determination of ADRs..The second goal in Phase II of this project is application of the new system for the prediction of ADRs for taxanes, which are anti-cancer compounds that target tubulin, which in turn forms microtubules. Microtubules are targets for numerous, therapeutically successful and chemically diverse anticancer drugs and are considered among the best targets for the anticancer compounds. The tubulin binding agents currently in clinical use include taxanes (paclitaxel and its analogs), vinblastine, vincristine, and vinorelbine. Our new system will be applied on taxanes for which we have in-house expertise (Chen et al., 2008; Huzil et al., 2007). Paclitaxel is routinely used to treat ovarian, breast and non-small cell lung cancer. We note that some of the ADRs are relatively well understood and will be used as true positives to validate the outputs generated by our new system..The ultimate goal of the Phase II is to find putative protein off-targets induced by the drug (paclitaxel). This would provide invaluable help in inferring or explaining the ADRs and, most importantly, would ultimately lead to improved drug design that maintains the high affinity interaction with the therapeutic target while eliminating the harmful interaction with the off-targets.