Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.

PARP-1 is an abundant nuclear enzyme that modifies substrates by poly(ADP-ribose)-ylation. PARP-1 has well-described functions in DNA damage repair and also functions as a context-specific regulator of transcription factors. With multiple models, data show that PARP-1 elicits protumorigenic effects in androgen receptor (AR)–positive prostate cancer cells, in both the presence and absence of genotoxic insult. Mechanistically, PARP-1 is recruited to sites of AR function, therein promoting AR occupancy and AR function. It was further confirmed in genetically defined systems that PARP-1 supports AR transcriptional function, and that in models of advanced prostate cancer, PARP-1 enzymatic activity is enhanced, further linking PARP-1 to AR activity and disease progression. In vivo analyses show that PARP-1 activity is required for AR function in xenograft tumors, as well as tumor cell growth in vivo and generation and maintenance of castration resistance. Finally, in a novel explant system of primary human tumors, targeting PARP-1 potently suppresses tumor cell proliferation. Collectively, these studies identify novel functions of PARP-1 in promoting disease progression, and ultimately suggest that the dual functions of PARP-1 can be targeted in human prostate cancer to suppress tumor growth and progression to castration resistance. Significance: These studies introduce a paradigm shift with regard to PARP-1 function in human malignancy, and suggest that the dual functions of PARP-1 in DNA damage repair and transcription factor regulation can be leveraged to suppress pathways critical for promalignant phenotypes in prostate cancer cells by modulation of the DNA damage response and hormone signaling pathways. The combined studies highlight the importance of dual PARP-1 function in malignancy and provide the basis for therapeutic targeting. Cancer Discov; 2(12); 1134–49. ©2012 AACR. This article is highlighted in the In This Issue feature, p. 1065

What's known on the subject? and What does the study add? Very few comparative studies to date evaluate the results of treatment options for prostate cancer using the most sensitive measurement tools. PSA has been identified as the most sensitive tool for measuring treatment effectiveness. To date, comprehensive unbiased reviews of all the current literature are limited for prostate cancer. This is the first large scale comprehensive review of the literature comparing risk stratified patients by treatment option and with long‐term follow‐up. The results of the studies are weighted, respecting the impact of larger studies on overall results. The study identified a lack of uniformity in reporting results amongst institutions and centres. A large number of studies have been conducted on the primary therapy of prostate cancer but very few randomized controlled trials have been conducted. The comparison of outcomes from individual studies involving surgery (radical prostatectomy or robotic radical prostatectomy), external beam radiation (EBRT) (conformal, intensity modulated radiotherapy, protons), brachytherapy, cryotherapy or high intensity focused ultrasound remains problematic due to the non‐uniformity of reporting results and the use of varied disease outcome endpoints. Technical advances in these treatments have also made long‐term comparisons difficult. The Prostate Cancer Results Study Group was formed to evaluate the comparative effectiveness of prostate cancer treatments. This international group conducted a comprehensive literature review to identify all studies involving treatment of localized prostate cancer published during 2000–2010. Over 18 000 papers were identified and a further selection was made based on the following key criteria: minimum/median follow‐up of 5 years; stratification into low‐, intermediate‐ and high‐risk groups; clinical and pathological staging; accepted standard definitions for prostate‐specific antigen failure; minimum patient number of 100 in each risk group (50 for high‐risk group). A statistical analysis (standard deviational ellipse) of the study outcomes suggested that, in terms of biochemical‐free progression, brachytherapy provides superior outcome in patients with low‐risk disease. For intermediate‐risk disease, the combination of EBRT and brachytherapy appears equivalent to brachytherapy alone. For high‐risk patients, combination therapies involving EBRT and brachytherapy plus or minus androgen deprivation therapy appear superior to more localized treatments such as seed implant alone, surgery alone or EBRT. It is anticipated that the study will assist physicians and patients in selecting treatment for men with newly diagnosed prostate cancer.