Abstract Clinical development of cancer therapies is associated with attrition rates as high as 80-95%. This high attrition suggests that standard preclinical pharmacology models do not accurately reflect clinical responses. The development of more predictive preclinical models requires several considerations; the relevance of the in vivo model, the administration of test agent, and the interpretation of efficacy data. PDX are cancer models developed from the direct transfer of patient tumor tissue into immunocompromised mice. A collection of PDX models, by retaining the genetic and histologic characteristics of the patients from which they were derived, represents the complexity and heterogeneity of human cancer. To minimize the clinical attrition rates of oncology compounds, we are developing hundreds of PDX models in seven major cancer indications. The collection is being molecularly profiled by RNAseq, WES, and proteomics. Profiling has identified models with robust expression of target proteins or mutant oncogenes that are likely to respond in preclinical efficacy tests. Conversely, the PDX models may provide an understanding of resistance, for example evaluating models with good target expression that fail to respond to therapy. Patient and tumor information, if known, has been collected for each PDX model including age, sex, cancer stage and grade, diagnosis, primary or metastatic site, and prior treatments. In addition to the improvements provided by the PDX models, a preclinical paradigm shift away from treatment with maximally tolerated dose towards clinically relevant dose (CRD), taking into consideration such aspects as exposure, formulation, route and schedule, is critical when attempting to predict clinical outcome from preclinical data. Also essential is the incorporation of clinically meaningful endpoints (regression) when assessing preclinical activity. We have initiated studies on cohorts of non small cell lung and breast PDX models to predict the likely clinical efficacy of candidate compounds for clinical development and to determine the CRD for standard of care (SOC) regimens required to define the most promising Phase II/III combination therapies. Anti-tumor activities were characterized using RECIST criteria of progressive disease (PD), stable disease (SD), partial response (PR), and complete response (CR). Target expression was evaluated by RNA, proteomics and immunohistochemistry. Preliminary results demonstrate a spectrum of responses against experimental therapeutics, including Phase I ADCs and are defining the CRD required for combination treatments with SOC. Identification of the most critical parameters of PDX models predicting clinical outcome will help in validating the utility of ‘n of 1′ studies with the PDX collection, inform patient enrollment strategies, guide combination therapies, and provide insight for identifying new tumor indications. Citation Format: Edward Rosfjord, Xin Han, Danielle Leahy, Erik Upeslacis, Justin Lucas, Jonathon Golas, Andrea Hooper, Fred Immermann, Bingwen Lu, Jeremy Myers, Zhengyan Kan, James Hardwick, Eric Powell, Puja Sapra, Paul Rejto, Hans-Peter Gerber, Judy Lucas. Patient derived xenograft (PDX) models: improving predictability of experimental cancer therapies. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 1469. doi:10.1158/1538-7445.AM2015-1469