Introduction Bipolar disorder (BD) is a chronic, disabling disease characterised by alternate mood episodes, switching through depressive and manic/hypomanic phases. Mood stabilizers, in particular lithium salts, constitute the cornerstone of the treatment in the acute phase as well as for the prevention of recurrences. The pathophysiology of BD and the mechanisms of action of mood stabilizers remain largely unknown but several pieces of evidence point to gene x environment interactions. Epigenetics, defined as the regulation of gene expression without genetic changes, could be the molecular substrate of these interactions. In this literature review, we summarize the main epigenetic findings associated with BD and response to mood stabilizers. Methods We searched PubMed, and Embase databases and classified the articles depending on the epigenetic mechanisms (DNA methylation, histone modifications and non-coding RNAs). Results We present the different epigenetic modifications associated with BD or with mood-stabilizers. The major reported mechanisms were DNA methylation, histone methylation and acetylation, and non-coding RNAs. Overall, the assessments are poorly harmonized and the results are more limited than in other psychiatric disorders (e.g. schizophrenia). However, the nature of BD and its treatment offer excellent opportunities for epigenetic research: clear impact of environmental factors, clinical variation between manic or depressive episodes resulting in possible identification of state and traits biomarkers, documented impact of mood-stabilizers on the epigenome. Conclusion Epigenetic is a growing and promising field in BD that may shed light on its pathophysiology or be useful as biomarkers of response to mood-stabilizer.

Autoencoders neural networks are nonlinear dimension reduction models widely used in the field of anomaly detection. Conventionally, the reconstruction error is considered as a score function allowing the discrimination between the normal data and the outliers. Recent advances in calculating uncertainty from neural networks open new perspectives in the field of anomaly detection. We study, for given models and different concentrations of anomalies, several score functions. We compare the standard score function based on the standard error, a score based on the error resulting from the Bayesian approximation, as well as score functions directly including the uncertainty. This paper empirically demonstrates how including uncertainty in the score function is likely to improve the performance of an autoencoder-based anomaly detection model.

No abstract