Improving the learning of chemical-protein interactions from literature using transfer learning and specialized word embeddings

In this paper, we explore the application of artificial neural network ('deep learning') methods to the problem of detecting chemical-protein interactions in PubMed abstracts. We present here a system using multiple Long Short Term Memory layers to analyse candidate interactions, to determine whether there is a relation and which type. A particular feature of our system is the use of unlabelled data, both to pre-train word embeddings and also pre-train LSTM layers in the neural network. On the BioCreative VI CHEMPROT test corpus, our system achieves an F score of 61.51% (56.10% precision, 67.84% recall).

X-exome sequencing of 405 unresolved families identifies seven novel intellectual disability genes

X-linked intellectual disability (XLID) is a clinically and genetically heterogeneous disorder. During the past two decades in excess of 100 X-chromosome ID genes have been identified. Yet, a large number of families mapping to the X-chromosome remained unresolved suggesting that more XLID genes or loci are yet to be identified. Here, we have investigated 405 unresolved families with XLID. We employed massively parallel sequencing of all X-chromosome exons in the index males. The majority of these males were previously tested negative for copy number variations and for mutations in a subset of known XLID genes by Sanger sequencing. In total, 745 X-chromosomal genes were screened. After stringent filtering, a total of 1297 non-recurrent exonic variants remained for prioritization. Co-segregation analysis of potential clinically relevant changes revealed that 80 families (20%) carried pathogenic variants in established XLID genes. In 19 families, we detected likely causative protein truncating and missense variants in 7 novel and validated XLID genes (CLCN4, CNKSR2, FRMPD4, KLHL15, LAS1L, RLIM and USP27X) and potentially deleterious variants in 2 novel candidate XLID genes (CDK16 and TAF1). We show that the CLCN4 and CNKSR2 variants impair protein functions as indicated by electrophysiological studies and altered differentiation of cultured primary neurons from Clcn4(-/-) mice or after mRNA knock-down. The newly identified and candidate XLID proteins belong to pathways and networks with established roles in cognitive function and intellectual disability in particular. We suggest that systematic sequencing of all X-chromosomal genes in a cohort of patients with genetic evidence for X-chromosome locus involvement may resolve up to 58% of Fragile X-negative cases.

The association between Polycystic Ovary Syndrome (PCOS) and metabolic syndrome: a statistical modelling approach

OBJECTIVE

Polycystic ovary syndrome (PCOS) affects 12-21% of women. Women with PCOS exhibit clustering of metabolic features. We applied rigorous statistical methods to further understand the interplay between PCOS and metabolic features including insulin resistance, obesity and androgen status.

DESIGN

Retrospective cross-sectional analysis.

PATIENTS

Women with PCOS attending reproductive endocrine clinics in South Australia for the treatment of PCOS (n = 172). Women without PCOS (controls) in the same Australian region (n = 335) from the Australian Diabetes, Obesity and Lifestyle Study (AusDiab), a national population-based study (age- and BMI-matched within one standard deviation of the PCOS cohort).

MEASUREMENTS

The factor structure for metabolic syndrome for women with PCOS and control groups was examined, specifically, the contribution of individual factors to metabolic syndrome and the association of hyperandrogenism with other metabolic factors.

RESULTS

Women with PCOS demonstrated clustering of metabolic features that was not observed in the control group. Metabolic syndrome in the PCOS cohort was strongly represented by obesity (standardized factor loading = 0·95, P < 0·001) and insulin resistance factors (loading = 0·92, P < 0·001) and moderately by blood pressure (loading = 0·62, P < 0·001) and lipid factors (loading = 0·67, P = 0·002). On further analysis, the insulin resistance factor strongly correlated with the obesity (r = 0·70, P < 0·001) and lipid factors (r = 0·68, P < 0·001) and moderately with the blood pressure factor (loading = 0·43, P = 0·002). The hyperandrogenism factor was moderately correlated with the insulin resistance factor (r = 0·38, P < 0·003), but did not correlate with any other metabolic factors.

CONCLUSIONS

PCOS women are more likely to display metabolic clustering in comparison with age- and BMI-matched control women. Obesity and insulin resistance, but not androgens, are independently and most strongly associated with metabolic syndrome in PCOS.

Obesity and polycystic ovary syndrome

Obesity is now a major international health concern. It is increasingly common in young women with reproductive, metabolic and psychological health impacts. Reproductive health impacts are often poorly appreciated and include polycystic ovary syndrome (PCOS), infertility and pregnancy complications. PCOS is the most common endocrine condition in women and is underpinned by hormonal disturbances including insulin resistance and hyperandrogenism. Obesity exacerbates hormonal and clinical features of PCOS and women with PCOS appear at higher risk of obesity, with multiple underlying mechanisms linking the conditions. Lifestyle intervention is first line in management of PCOS to both prevent weight gain and induce weight loss; however improved engagement and sustainability remain challenges with the need for more research. Medications like metformin, orlistat, GLP1 agonists and bariatric surgery have been used with the need for large scale randomised clinical trials to define their roles.

Risk stratification in early pregnancy for women at increased risk of gestational diabetes

AIM

To evaluate the addition of fasting glucose and lipids to a simple, validated risk prediction tool for gestational diabetes (GDM) applied in early pregnancy.

METHODS

Women at risk of developing GDM on a validated risk prediction tool were recruited in early pregnancy into a large randomised controlled trial. Outcome measures included fasting biochemical markers (glucose, lipids) at 12-15 weeks gestation and GDM diagnosis (28 weeks gestation). Multivariable logistic regression was used to identify additional predictive biochemical variables for GDM, with corresponding receiver operator characteristic (ROC) curves generated. Unadjusted and adjusted models were derived for both the Australasian Diabetes in Pregnancy (ADIPS) and the International Association for Diabetes in Pregnancy Study Group (IADPSG) GDM diagnostic criteria.

RESULTS

51 (23%) Women were diagnosed with GDM based on ADIPS criteria, with 60 (30%) diagnosed based on IADPSG criteria. In all four regression models, fasting glucose was the strongest predictor for GDM development with an odds ratio range of 4.7-6.3 (ADIPS) and 8.8-10 (IADPSG). ROC curves revealed an area under the curve of 0.79 (95% CI: 0.72-0.86) for ADIPS criteria and 0.83 (95% CI: 0.77-0.90) for IADPSG criteria for adjusted models.

CONCLUSIONS

In a two-step approach, when applied with a validated risk prediction tool, fasting glucose in early pregnancy was predictive of GDM and incrementally improved risk identification, presenting potential for an early pregnancy, GDM risk screening strategy for streamlining of pregnancy care and opportunity for preventive intervention.

The mechanics of PLGA nanofiber scaffolds with biomimetic gradients in mineral for tendon-to-bone repair

Attachment of dissimilar materials is prone to failure due to stress concentrations that can arise their interface. A compositionally or structurally graded transition can dissipate these stress concentrations and thereby toughen an attachment. The interface between compliant tendon and stiff bone utilizes a monotonic change in hydroxylapatite mineral ("mineral") content to produce a gradient in mechanical properties and mitigate stress concentrations. Previous efforts to mimic the natural tendon-to-bone attachment have included electrospun nanofibrous polymer scaffolds with gradients in mineral. Mineralization of the nanofiber scaffolds has typically been achieved using simulated body fluid (SBF). Depending on the specific formulation of SBF, mineral morphologies ranged from densely packed small crystals to platelike crystal florets. Although this mineralization of scaffolds produced increases in modulus, the peak modulus achieved remained significantly below that of bone. Missing from these prior empirical approaches was insight into the effect of mineral morphology on scaffold mechanics and on the potential for the approach to ultimately achieve moduli approaching that of bone. Here, we applied two mineralization methods to generate scaffolds with spatial gradations in mineral content, and developed methods to quantify the stiffening effects and evaluate them in the context of theoretical bounds. We asked whether either of the mineralization methods we developed holds potential to achieve adequate stiffening of the scaffold, and tested the hypothesis that the smoother, denser mineral coating could attain more potent stiffening effects. Testing this hypothesis required development of and comparison to homogenization bounds, and development of techniques to estimate mineral volume fractions and spatial gradations in modulus. For both mineralization strategies, energy dispersive X-ray analysis demonstrated the formation of linear gradients in mineral concentration along the length of the scaffolds, and Raman spectroscopic analysis revealed that the mineral produced was hydroxylapatite. Mechanical testing showed that the stiffness gradient using the new method was significantly steeper. By analyzing the scaffolds using micromechanical modeling techniques and extrapolating from our experimental results, we present evidence that the new mineralization protocol has the potential to achieve levels of stiffness adequate to contribute to enhanced repair of tendon-to-bone attachments.

Can the fetal fibronectin test be used by remote dwelling pregnant women to predict the onset of labour at term and delay transfer for birth in regional settings?

INTRODUCTION

The closure of rural maternity units in Australia means an increasing number of women are transferred into major centres to await birth. Accurately excluding the onset of labour could delay relocation. The fetal fibronectin (fFN) test is used to predict preterm birth; however, the accuracy of this test for determining impending term birth is unclear.

METHODS

75 women were recruited to this study from<strong> </strong>two remote maternity units. Eligibility criteria were: aged &&num;8805;18 years, singleton pregnancy, 37+0-40+3 weeks (37 weeks to 40 weeks and 3 days gestation) and no indication for induction of labour or caesarean section in next 7 days. The Quikcheck fFN® test was performed at 37 weeks and then repeated at 7 day intervals. Time-to-birth from test date was modelled using linear regression. Logistic regression models estimated odds of birth within 7 days. Separate models considered first and last test results and those at 38 weeks; adjusted for use of lubricant and gestational age.

RESULTS

A shorter time-to-birth was found in women with positive compared with negative fFN tests; significant at first fFN test (adjusted mean difference [AMD] 5.4 days, 95% CI 2.0-8.8) and 38 weeks (AMD 5.7 days, 95% CI 2.2-9.2 days). A positive test was also associated with a significant increase in the odds of birthing within 7 days: first fFN test adjusted odds ratio (AOR) 11.0 (95% CI 2.5-48.7), 38 weeks test AOR 14.4 (95% CI 3.4-60.2), last fFN test AOR 8.1 (95% CI 1.6-39.8). However, of women who gave birth within 7 days of testing a significant proportion had a negative fFN result; first fFN test 8/17(47.1%), 38 weeks test 4/14(28.6%) and last fFN test 29/58(50.0%).

CONCLUSION

The presence of fFN in cervical secretions was associated with impending term birth but its absence did not reliably exclude the onset of birth. Delaying transfer based on these findings would result in some women birthing in their home communities.