DescriptionTogether oral bioavailability and hepatotoxicity determine the fate and failure of a new drug in clinical trials. A promising drug candidate that has little to no oral bioavailability could be considered an ineffective treatment. If the drug does have high oral bioavailability, yet exhibits severe hepatotoxicity, the drug will be withdrawn from clinical trials. Thus, oral bioavailability and hepatotoxicity account for a substantial number of drugs eliminated from therapies and withdrawn from the market. Not only is this a severe financial loss for a pharmaceutical company; it is also a loss for patients who will no longer be able to benefit from the therapeutic effects. The most common approach to testing oral bioavailability and hepatotoxicity before clinical trials is through animal testing. The information learned from animal testing is highly valuable, but is expensive, time consuming, and has low throughput. Cell based assays developed to study specific biochemical mechanisms and toxicity are heavily used as an alternative; however, correlations between complex in vitro and in vivo endpoints are not very clear. Another alternative is Quantitative Structure Activity Relationship (QSAR) approach. A QSAR model could evaluate millions of chemicals without requiring them to be synthesized, which saves money, time, and has high throughput. Many predictive QSAR models have been developed since the 1960’s, yet predictive in vivo QSAR models are difficult to build and rare to find. Developing new methods to integrate mechanistic information and improve in vitro-in vivo correlation(s) for QSAR modeling purposes have been shown to be beneficial and are the focus of this thesis. The complex in vivo endpoints oral bioavailability and hepatotoxicity were used as example endpoints to model. Methods to curate biological information and in vitro data, specifically high-throughput screening data, and incorporate it into the QSAR models are discussed in great detail. The performance of the resulting models, as well as their shortcomings is also discussed. Overall, incorporating biological information into the QSAR modeling workflow greatly improved predictions from both the oral bioavailability and hepatotoxicity models. The new techniques can be adapted to model other complex biological endpoints.