We propose a model-based metric to estimate the factual accuracy of generated text that is complementary to typical scoring schemes like ROUGE (Recall-Oriented Understudy for Gisting Evaluation) and BLEU (Bilingual Evaluation Understudy). We introduce and release a new large-scale dataset based on Wikipedia and Wikidata to train relation classifiers and end-to-end fact extraction models. The end-to-end models are shown to be able to extract complete sets of facts from datasets with full pages of text. We then analyse multiple models that estimate factual accuracy on a Wikipedia text summarization task, and show their efficacy compared to ROUGE and other model-free variants by conducting a human evaluation study.

Clostridium difficile, a major cause of hospital-acquired diarrhea, triggers disease through the release of two toxins, toxin A (TcdA) and toxin B (TcdB). These toxins disrupt the cytoskeleton of the intestinal epithelial cell, increasing intestinal permeability and triggering the release of inflammatory mediators resulting in intestinal injury and inflammation. The most prevalent animal model to study TcdA/TcdB-induced intestinal injury involves injecting toxin into the lumen of a surgically generated "ileal loop." This model is time-consuming and exhibits variability depending on the expertise of the surgeon. Furthermore, the target organ of C. difficile infection (CDI) in humans is the colon, not the ileum. In the current study, we describe a new model of CDI that involves intrarectal instillation of TcdA/TcdB into the mouse colon. The administration of TcdA/TcdB triggered colonic inflammation and neutrophil and macrophage infiltration as well as increased epithelial barrier permeability and intestinal epithelial cell death. The damage and inflammation triggered by TcdA/TcdB isolates from the VPI and 630 strains correlated with the concentration of TcdA and TcdB produced. TcdA/TcdB exposure increased the expression of a number of inflammatory mediators associated with human CDI, including interleukin-6 (IL-6), gamma interferon (IFN-␥), and IL-1␤. Finally, we were able to demonstrate that TcdA was much more potent at inducing colonic injury than was TcdB but TcdB could act synergistically with TcdA to exacerbate injury. Taken together, our data indicate that the intrarectal murine model provides a robust and efficient system to examine the effects of TcdA/TcdB on the induction of inflammation and colonic tissue damage in the context of human CDI.

As the current efficacy of oncolytic viruses (OVs) as monotherapy is limited, exploration of OVs as part of a broader immunotherapeutic treatment strategy for cancer is necessary. Here, we investigated the ability for immune checkpoint blockade to enhance the efficacy of oncolytic reovirus (RV) for the treatment of breast cancer (BrCa). In vitro, oncolysis and cytokine production were assessed in human and murine BrCa cell lines following RV exposure. Furthermore, RV-induced upregulation of tumor cell PD-L1 was evaluated. In vivo, the immunocompetent, syngeneic EMT6 murine model of BrCa was employed to determine therapeutic and tumor-specific immune responses following treatment with RV, anti-PD-1 antibodies or in combination. RV-mediated oncolysis and cytokine production were observed following BrCa cell infection and RV upregulated tumor cell expression of PD-L1. In vivo, RV monotherapy significantly reduced disease burden and enhanced survival in treated mice, and was further enhanced by PD-1 blockade. RV therapy increased the number of intratumoral regulatory T cells, which was reversed by the addition of PD-1 blockade. Finally, dual treatment led to the generation of a systemic adaptive anti-tumor immune response evidenced by an increase in tumor-specific IFN-γ producing CD8+ T cells, and immunity from tumor re-challenge. The combination of PD-1 blockade and RV appears to be an efficacious immunotherapeutic strategy for the treatment of BrCa, and warrants further investigation in early-phase clinical trials.