Catherine Mooney

Background: Major Depressive Disorder (MDD) is one of the leading causes of disability worldwide. Deep learning methods have been widely used for MDD detection, with research suggesting that deep models outperform traditional machine learning techniques. However, detecting MDD remains challenging due to data heterogeneity, model complexities and the requirement for discriminative feature representations. Objective: This review outlines recent progress in deep learning methods for MDD detection from Resting-state fMRI (rs-fMRI), with a focus on the model’s generalisability and features that most effectively represent the function/anatomy of the brain to contribute to biomarker identifications and interpretability. Further, the review assesses the applicability of current models to real-world challenges. Methods: This systematic review followed the PRISMA guidelines. Studies involved clinically diagnosed MDD subjects, a control group, and deep learning methods for classification tasks. Results: The cerebellum, thalamus, amygdala, insula, and default mode network are the most frequently reported brain regions associated with depression. Although deep learning has shown impressive results, it has limitations in terms of reliance on labelled data, heterogeneity of data from various hospitals, and model interpretability. A majority of the studies lacked external validation and had a single-site dataset or regionally homogeneous datasets, and did not consider the temporal and dynamic nature of rs-fMRI data. Conclusion: Deep learning offers considerable potential in advancing MDD diagnosis and understanding its mechanisms. Multi-regional data collection, harmonisation techniques, and rigorous testing in real-world workflows should be the primary focus of future research.

Electroencephalogram (EEG) has been the gold standard for measuring epileptic activity in rodent models of epilepsy. Manual scoring of seizures in EEG recordings lasting from days to months is laborious and prone to human error. The existing literature on automatic seizure detection in rodent models of epilepsy is limited, and the electrographic characteristics of induced epilepsy significantly differ from those of other epilepsy types. This study employed a Light Gradient Boosting Machine (LightGBM), with the dataset carefully partitioned into separate training and testing sets to ensure no data overlap. The model was trained using five-fold cross-validation to enhance robustness and generalisability. The training, validation, and independent test sets comprised 29,722 h of EEG recordings from 102 mice with pilocarpine-induced temporal lobe epilepsy. Following feature selection, model training, and post-processing, the lightGBM-based model exhibited a sensitivity of 80%, a specificity of 99%, and an F1-score of 0.71 on the independent test set. Multiple pairwise and non-parametric statistical tests indicated that envelope, skewness, and kurtosis, identified as the three most significant features in the feature importance ranking, exhibit statistically significant differences in their distributions (p-value < 0.05). The statistical analysis revealed significant differences across the three features and between seizure and non-seizure events for each feature, highlighting their relevance for discriminating epileptic activity. This study highlights the potential to support the automation of seizure event detection in preclinical rodent models of epilepsy.

MicroRNAs (miRNA) are small non-coding RNAs that are key negative regulators of gene expression. Their roles include shaping the gene expression landscape during and after brain development by defining and maintaining levels of proteins that generate the distinct morphological and functional properties of neurons and other brain cell types. HT22, N2A, and SH-SY5Y are common immortalized neuronal cell lines that offer simple, less expensive, and time-saving options for in vitro modelling to evaluate miRNA functions. The extent to which these lines reflect primary neurons remains, however, unclear. Here, we benchmarked the miRNA profiles of cultured mouse hippocampal neurons against Argonaute-loaded miRNAs in the adult mouse hippocampus and miRNA data from the hippocampus of control human donors. We then compared the miRNA expression landscape in HT22, N2A and SH-SY5Y against mouse hippocampal primary cell cultures. We profiled over 700 miRNAs across the lines and detected 310 miRNAs in all four cell types. This included detection of neuron-enriched miRNAs such as miR-124 and miR-128, although the cell lines typically displayed lower levels of these than in primary neurons and reference adult hippocampal tissue. The miRNA profile in the HT22 cell line showed the highest correlation to the mouse primary neuronal cultures. Together, this study provides a dataset on basal miRNA expression across commonly used cell lines for neuroscience research and evidence for both conserved and distinct profiles that should be used to inform decisions on cell lines for modelling brain and miRNA research.

Background: Major Depressive Disorder (MDD) is one of the leading causes of disability, and it is currently diagnosed solely based on self-reported symptoms. Functional magnetic resonance imaging (fMRI) provides information about both brain activity and anatomy, making it a promising tool for detecting depression. However, deep learning models have difficulties in generalising and are highly reliant on available labelled data. Objectives: This study proposes a deep learning approach for detecting MDD from fMRI data with the goal of enhancing model generalisation across different sites, which has often been overlooked in previous studies. Methods: The dataset consisted of 89 subjects with depression and 63 healthy controls. Functional Connectivity (FC) was used as the input feature. First, a Vision Transformer (ViT) was applied for classification. Then, a Siamese network, consisting of two identical ViT backbones, was implemented using contrastive learning to improve feature representation. The weights of the pretrained Siamese network were used for fine-tuning the ViT model. Three Independent datasets obtained from multiple regions were also used for further assessing the model’s generaliseability. Results: Siamese ViT achieved 87.5% accuracy and demonstrated improved generalisation across independent and cross-regional test sets. It significantly enhanced recall and reduced false negatives. Cerebellar lobules were identified as consistent discriminative regions. Conclusion: Siamese transfer learning outperformed baseline ViT, demonstrating that pretraining networks enhance feature extraction. The contrastive learning approach, by comparing two inputs simultaneously, captures nuanced brain activity patterns and has great potential in reducing the reliance on annotated data.

Background: Individuals affected by psychotic disorders often decide to have a trial of dose reduction or discontinuation of their antipsychotic medications; however, there is a lack of evidence-based guidance for this practice. There is limited knowledge on the different methods for dose reduction or discontinuation of antipsychotic medications in which methods might be the safest and most effective. This scoping review aims to identify methods of antipsychotic dose reduction and discontinuation, to develop a comprehensive inventory of tapering strategies. Study Design: A systematic search was performed in PubMed, EMBASE, PsycINFO, and Cochrane to identify publications involving antipsychotic dose reduction or discontinuation. We included primary studies, systematic reviews, theoretical papers, and guidelines. Study Results: A total of 9005 unique articles were identified, and full text screening was conducted on 459 of these. Thirty-two articles were included in the final analysis: 28 related to oral antipsychotic medications and 4 to long acting injectables. Three broad categories of dose reduction/discontinuation were identified—immediate, rapid (<4 weeks), and gradual (short 4-26 weeks and long≥26 weeks). Further subtypes were identified, with reduction based on either the original starting dose or the most recent dose after the last reduction. The hyperbolic method estimates the dose reduction based on a gradual reduction in the D2 receptor occupancy. Conclusions: A wide range of antipsychotic tapering methods have been implemented internationally with no consensus on the best approach. The results of this scoping review could help to inform research on establishing the most effective methods of antipsychotic discontinuation.

There is considerable evidence implicating maternal immune activation (MIA) and cytokine dysregulation in the pathophysiology of Autism. However, cytokines, due to their lack of specificity are unlikely to translate clinically as prognostic biomarkers. Our aim was to explore the perinatal molecular pathways dysregulated in umbilical cord blood, which precede a diagnosis of childhood Autism, and ascertain whether these putative biomarkers persisted into pre-pubertal childhood. In a cohort of 2137 mother-infant dyads, we conducted a nested case-control study in the BASELINE Birth Cohort. Proteomic and metabolomic analysis was performed on cord blood plasma from 22 children diagnosed with Autism before age 5, and 44 neurotypical controls. In a clinical diagnostic follow-up between 7–10 years in the PiRAMiD Cohort, 24 children with Autism and 48 controls provided blood samples for molecular profiling. In cord blood, proteomics revealed altered glycolysis, selenium metabolism, oxygen transport, and complement signalling. Alterations in these protein pathways persisted into childhood, and dysregulation of GAPDH, SELENBP1, and BLVRB proteins were evident in both cord blood and in serum from pre-pubertal children with Autism. In cord blood, metabolomics analysis indicated Autism outcome was associated with reduced levels of circulating steroids and increased sulfate. We confirmed androstenedione was reduced in cord blood, in Autism cases in comparison to controls, however changes in androstenedione levels were not evident in serum from pre-pubertal children with Autism. Our findings were further corroborated using machine learning approaches, with an AUROC ranging from 0.82 to 0.85 for proteomic and metabolomic cord blood prediction models, respectively. Collectively, these findings confirm a cord blood molecular signature precedes the onset of Autism and has the potential to lead to prognostic biomarkers. Our integrative multi-omics analysis reveals materno-feto-placental molecular processes which potentially underpin Autism aetiology.