Abstract Tumor molecular profiling is rapidly becoming the standard clinical test for selecting targeted therapies in refractory cancer patients. DNA extracted from patient samples is enriched for cancer genes and sequenced to identify actionable somatic mutations therein. A major challenge arises when tumor-derived data is analyzed in the absence of normal tissue data, as it is common in clinical scenarios. The distinction between somatic and germline variants become difficult, leaving clinicians to resort to crude heuristic filtering. We present here a variant calling software, developed under quality system regulation protocols, capable of accurately identifying somatic mutations from targeted next-generation sequencing data. A novel Bayesian Network approach models the distribution of reads harboring germline and somatic mutations, estimates the contamination from normal tissue in the sample, scores somatic mutations, and imputes germline variants, without matching normal tissue data. This approach also allows joint analysis of multiple specimens from the same patient (e.g. FFPE and ctDNA), when available, improving the limit of detection. To improve specificity, our caller can also utilize prior information from different databases including somatic mutations, germline variation, and healthy controls data, in a principled fashion. We validated our method by analyzing data from the TOMA OS-Seq 131 cancer gene panel using the Illumina platform. Sample inputs ranging from 2-600ng of DNA were sequenced to a depth of >1000X, achieving on target rates ≤73% and uniformity ≥ 3.2 fold 80 penalty. Through adaptors with molecular barcodes we measured a median duplicate rate <2. We analyzed somatic mutations simulated at various variant allele fractions on a background of data from reference samples from the Genome-in-a-Bottle consortium, data on a dilution series from two reference samples, and several commercial control and clinical samples, including matched FFPE, PBMC, and ctDNA specimens. In the absence of normal tissue, our method scores each variant with respect to their likelihood of being somatic or germline. We show that, as compared to other commonly used methods, our algorithm can achieve a higher true positive rate whilst controlling a false discovery rate of 1%. We also show that jointly analyzing serial samples (e.g. ctDNA), we can improve sensitivity of shared variants. In conclusion, in contrast to currently used academic software developed for research projects, we observe that our caller outperforms these software and is particularly well suited for the clinical use cases. Note: This abstract was not presented at the meeting. Citation Format: Francisco M. De La Vega, Sean Irvine, David Ware, Kurt Gaastra, Yannick Pouiliot, Len Trigg. Accurate identification of somatic mutations in cancer patient specimens in the lack of normal tissue by targeted high-throughput sequencing [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 3576. doi:10.1158/1538-7445.AM2017-3576