Abstract Introduction: Cancers are diverse diseases reflective of tissue (histopathology) and demographic origins, molecular pathogenesis, and pharmacology. Patient-derived-xenograft (PDX) represents an experimental system most closely mimicking these features of original patients. The following parameters are critical for making PDX productive and valuable: 1) a large library of PDX reflective of disease diversity; 2) accurate pathology diagnosis and patient information being annotated and QC’d; 3) ideally, also including patient treatments along with follow-up; 4) having full corresponding genetic/genomic profile (molecular pathology); 5) capacity to conduct large-scale mouse clinical trial (MCT); and 6) generating and curating a large collection of experimental datasets (e.g., growth kinetics, drug responses). Methods: We built a large collection of PDXs from different patient populations of disease types {An, 20107; Chen, 2015; Guo, 2016; Yang, 2016; Zhang, 2013) and demographics in our research sites in China, US, and UK. We also built a database (HuBase®) to curate, store, QC, and integrate all the above information, which is readily accessible for researchers. We have also built operations to conduct large MCT at the sites. Results: We have built the world's largest PDX library that now contains ~3,000 PDXs from both Asian (~half) and Western (~half) patients, covering ~30 major cancer types, including some rare/difficult ones to establish (http://crownbio.com/eortc17/stromalposter). The large disease panels (each have 100~400) include NSCLC, CRC, gastric, pancreatic, and HCC (Chen, 2015; Guo, 2016; Yang, 2013; Zhang, 2013); the medium ones with each >50 include esophageal, H&N, ovarian, cholangiocarcinoma, etc.; and small rare/difficult panels include ALL, AML (An, 2017), lymphoma, TNBC, GIST, GBM, etc. The large/medium panels are particularly meaningful to support MCT (Chen, 2015 #815; Zhang, 2013 #525). All models are under characterization/annotation systematically, while a large subset (approaching 2,000) with meaningful annotations has been searchable in HuBase®. Importantly, ~50% of our collection (~1,500) has been genomic-profiled using different technology platforms, earlier by various microarrays (SNP6.0, GenChip U133/U219) and now by NGS (RNAseq/WES). SNP6.0 then, while WES now by proprietary algorithm, characterizes GCN; RNAseq now (U219/133 then) characterizes expression; WES/RNAseq profile mutations and RNAseq also profile gene fusion. These, along with the efficacy data, enable powerful biomarker/drug target discovery/validation. All models are QC’d by STR/HLA for ID and molecular-pathology QC’d by proprietary algorithm (Guo, 2016). Many driver mutations/specific drug response/resistance have been identified and validated, enabling specific target treatment evaluations. HLA typing also enables I/O research. Our database also tracks live status and locations of the models, enabling timely and multicenter MCT. Conclusions: Our PDX platform is proven useful for cancer search and evaluating candidate treatments, including discovery/validation of predictive biomarkers and drug targets. Citation Format: Jie Cai, Xuesong Ouyang, Sheng Guo, Likun Zhang, Jessie Wang, Wubin Qian, Jia Xue, Bin Fan, Jayant Thatte, Rajendra Kumari, Henry Qixiang Li. In-depth annotations of the world's largest patient tumor-derived xenogragft (PDX) library for cancer research and preclinical pharmacology evaluation [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2017 Oct 26-30; Philadelphia, PA. Philadelphia (PA): AACR; Mol Cancer Ther 2018;17(1 Suppl):Abstract nr A013.