(595b) In Silico Prediction of Cancer GI50

Current methods of cancer treatment have improved vastly the survivability of humans from many forms of cancer. Traditional chemotherapy remains a key treatment method, offering distinct advantages over other treatment options, especially in the treatment of metastasized tumors. Much effort has been devoted to identifying new, more effective chemotherapeutic agents. Currently, the National Cancer Institute (NCI) analyzes the potential anticancer activity of over 3,000 compounds per year. However, less than one percent of these drugs will be marketed, mostly due to poor anticancer activity and toxicity issues.

To provide a faster and more accurate methodology for identification of potential anticancer agents, we have developed in silico QSAR (quantitative structure-activity relationship) models to predict the concentration of chemical required to reduce the cancer growth rate by 50% (GI₅₀) for the breast cancer cell line MCF-7. These screening models use nonlinear neural networks to provide a priori predictions of GI₅₀ for potential anticancer agents with an overall root-mean square error of 0.49 Log (GI₅₀). As a screening tool capable of greatly reducing time and expense of new drug development, the model is employed to remove unqualified candidate anticancer compounds, while identifying and retaining highly qualified candidates for experimental trials. Further, this methodology facilitates a greater understanding of the relationship between molecular structure and growth inhibition potential.