PhosphoDisco: A Toolkit for Co-regulated Phosphorylation Module Discovery in Phosphoproteomic Data.

Kinases are key players in cancer-relevant pathways and are the targets of many successful precision cancer therapies. Phosphoproteomics is a powerful approach to study kinase activity and has been used increasingly for the characterization of tumor samples leading to the identification of novel chemotherapeutic targets and biomarkers. Finding co-regulated phosphorylation sites which represent potential kinase-substrate sets or members of the same signaling pathway allows us to harness these data to identify clinically relevant and targetable alterations in signaling cascades. Unfortunately, studies have found that databases of co-regulated phosphorylation sites are only experimentally supported in a small number of substrate sets. To address the inherent challenge of defining co-regulated phosphorylation modules relevant to a given dataset, we developed PhosphoDisco, a toolkit for determining co-regulated phosphorylation modules. We applied this approach to tandem mass spectrometry based phosphoproteomic data for breast and non-small cell lung cancer and identified canonical as well as putative new phosphorylation site modules. Our analysis identified several interesting modules in each cohort. Among these was a new cell cycle checkpoint module enriched in basal breast cancer samples and a module of PRKC isozymes putatively co-regulated by CDK12 in lung cancer. We demonstrate that modules defined by PhosphoDisco can be used to further personalized cancer treatment strategies by establishing active signaling pathways in a given patient tumor or set of tumors, and in providing new ways to classify tumors based on signaling activity.

Correlation between perioperative outcomes and long-term survival for non-small lung cancer treated at major centers.

BACKGROUND: The public is placing increased emphasis on specialty specific rankings, thereby affecting patients' choices of clinical care programs. In the spirit of transparency, public reporting initiatives are underway or being considered by various surgical specialties whose databases rank programs based on short-term outcomes. Of concern, short-term risk avoidance excludes important comparative cases from surgical database participation and may adversely affect overall long-term oncologic treatment team results. To assess the validity of comparing short-term perioperative and long-term survival outcomes of all patients treated at major centers, we studied the correlations between these variables. METHODS: The National Cancer Database was queried for patients diagnosed with non-small cell lung carcinoma (NSCLC) between 2008 and 2012, yielding 5-year follow-up data for all patients at centers treating at least 100 patients annually. Mortality (30- and 90-day), unplanned 30-day readmissions, and hospital length of stay were modeled using logistic regression with sex, race, age, Charlson-Deyo combined comorbidity, extent of surgery, income, insurance status, histology, grade, and analytic stage as predictors, all with 2-way interaction terms. The differences between the predicted rates and observed rates were calculated for each short-term outcome, and the average of these was used to create a short-term metric (STM). A similar approach was used to create a long-term metric (LTM) that used overall survival as a single dependent variable. Centers were ranked into deciles based on these metrics. Visual plotting as well as correlation coefficients were used to judge correlation between STM and LTM. RESULTS: A total of 298,175 patients from 541 centers were included in this analysis, of whom 102,860 underwent surgical resection for NSCLC. The correlation between STM and LTM was negative using parametric estimates (Pearson correlation coefficient = -0.09 [P = .03] and -0.22 [P < .01]) and nonparametric estimates (Spearman rank correlation coefficient = -0.09 [P = .02] and -0.22 [P < .01]) for squamous cell carcinoma and adenocarcinoma, respectively. CONCLUSIONS: Short-term perioperative outcome rankings correlate poorly with long-term survival outcome rankings when cancer treatment centers are compared. Factors explaining this discrepancy merit further study. Rankings based on short-term outcomes alone may be incomplete for public reporting.