Publications

Gi-GPCRs, G protein-coupled receptors that signal via Gα proteins of the i/o class (Gαi/o), acutely regulate cellular behaviors widely in mammalian tissues, but their impact on the development and growth of these tissues is less clear. For example, Gi-GPCRs acutely regulate insulin release from pancreatic β cells, and variants in genes encoding several Gi-GPCRs--including the α-2a adrenergic receptor, ADRA2A--increase the risk of type 2 diabetes mellitus. However, type 2 diabetes also is associated with reduced total β-cell mass, and the role of Gi-GPCRs in establishing β-cell mass is unknown. Therefore, we asked whether Gi-GPCR signaling regulates β-cell mass. Here we show that Gi-GPCRs limit the proliferation of the insulin-producing pancreatic β cells and especially their expansion during the critical perinatal period. Increased Gi-GPCR activity in perinatal β cells decreased β-cell proliferation, reduced adult β-cell mass, and impaired glucose homeostasis. In contrast, Gi-GPCR inhibition enhanced perinatal β-cell proliferation, increased adult β-cell mass, and improved glucose homeostasis. Transcriptome analysis detected the expression of multiple Gi-GPCRs in developing and adult β cells, and gene-deletion experiments identified ADRA2A as a key Gi-GPCR regulator of β-cell replication. These studies link Gi-GPCR signaling to β-cell mass and diabetes risk and identify it as a potential target for therapies to protect and increase β-cell mass in patients with diabetes.

While significant effort has been dedicated to the characterization of epigenetic changes associated with prenatal differentiation, relatively little is known about the epigenetic changes that accompany post-natal differentiation where fully functional differentiated cell types with limited lifespans arise. Here we sought to address this gap by generating epigenomic and transcriptional profiles from primary human breast cell types isolated from disease-free human subjects. From these data we define a comprehensive human breast transcriptional network, including a set of myoepithelial- and luminal epithelial-specific intronic retention events. Intersection of epigenetic states with RNA expression from distinct breast epithelium lineages demonstrates that mCpG provides a stable record of exonic and intronic usage, whereas H3K36me3 is dynamic. We find a striking asymmetry in epigenomic reprogramming between luminal and myoepithelial cell types, with the genomes of luminal cells harbouring more than twice the number of hypomethylated enhancer elements compared with myoepithelial cells.

BACKGROUND AND STUDY AIMS

The real-time optical diagnosis of colorectal polyps with high confidence predictions can achieve high levels of accuracy. Increasing the rates of high confidence optical diagnosis can improve the clinical application of real-time optical diagnosis in routine practice. The primary aim of this prospective study was to evaluate whether high magnifying endoscopy improves the rates of high confidence narrow-band imaging (NBI) - based optical diagnosis for differentiating between neoplastic and non-neoplastic colorectal lesions according to the NBI international colorectal endoscopic (NICE) classification.

PATIENTS AND METHODS

Consecutive adult patients undergoing colonoscopy with a high magnifying (maximum, × 80) colonoscope between April and August 2012 were recruited. The optical diagnosis for each polyp was evaluated during colonoscopy in two consecutive stages by the same endoscopist, who first used NBI with non-magnifying endoscopy (NBI-NME), then NBI with magnifying endoscopy (NBI-ME). A level of confidence was assigned to each prediction.

RESULTS

The analysis included 124 patients (mean age, 56.4 years; male-to-female ratio, 72:52) with 248 polyps smaller than 10 mm. Of the 248 polyps, 210 were 1 to 5 mm in size and 38 were 6 to 9 mm in size; 77 polyps were hyperplastic, 4 were sessile serrated adenomas/polyps, 160 were low grade adenomas, 5 were high grade adenomas, and 2 were deep submucosal invasive carcinomas. The rate of high confidence optical diagnosis when NBI-ME was used was significantly higher than the rate when NBI-NME was used for diminutive (1 - 5 mm) polyps (92.9 % vs 79.5 %, P < 0.001) and for small (6 - 9 mm) polyps (94.7 % vs 84.2 %, P = 0.048).

CONCLUSION

High magnifying endoscopy significantly improved the rates of high confidence NBI-based optical diagnosis of diminutive and small colorectal polyps.

STUDY REGISTRATION

UMIN 000007608.

BACKGROUND AND SIGNIFICANCE

Sparsity is often a desirable property of statistical models, and various feature selection methods exist so as to yield sparser and interpretable models. However, their application to biomedical text classification, particularly to mortality risk stratification among intensive care unit (ICU) patients, has not been thoroughly studied.

OBJECTIVE

To develop and characterize sparse classifiers based on the free text of nursing notes in order to predict ICU mortality risk and to discover text features most strongly associated with mortality.

METHODS

We selected nursing notes from the first 24h of ICU admission for 25,826 adult ICU patients from the MIMIC-II database. We then developed a pair of stochastic gradient descent-based classifiers with elastic-net regularization. We also studied the performance-sparsity tradeoffs of both classifiers as their regularization parameters were varied.

RESULTS

The best-performing classifier achieved a 10-fold cross-validated AUC of 0.897 under the log loss function and full L2 regularization, while full L1 regularization used just 0.00025% of candidate input features and resulted in an AUC of 0.889. Using the log loss (range of AUCs 0.889-0.897) yielded better performance compared to the hinge loss (0.850-0.876), but the latter yielded even sparser models.

DISCUSSION

Most features selected by both classifiers appear clinically relevant and correspond to predictors already present in existing ICU mortality models. The sparser classifiers were also able to discover a number of informative - albeit nonclinical - features.

CONCLUSION

The elastic-net-regularized classifiers perform reasonably well and are capable of reducing the number of features required by over a thousandfold, with only a modest impact on performance.