Abdalrhman Mostafa

ID 20324 Poster Board 99 Machine learning (ML) models have demonstrated impressive applicability in classifying hemodynamic signals according to disease-specific features of overt cardiovascular (CV) conditions like myocardial infarction and hypertension. However, there remains a pressing need for identifying subtle, subclinical deviations in hemodynamic time-series associated with early disease states, to predict prognosis and prevent deteriorating processes. Our previous work indicates the predictive capacity of non-linear parameters of blood pressure variability in discriminating between healthy and prediabetic rats in a sex- and age-specific manner. Whilst pre-diabetic rats lack signs of gross CV insult, they possess blunted baroreceptor sensitivity and dysfunctional endothelial-dependent vasorelaxation only discernible upon complex, invasive hemodynamic manipulations. Our results show that female rats are relatively protected from prediabetic CV manifestations. Here, we evaluate the capacity of a ML convolutional neural network (CNN) in segregating arterial pressure (AP) signals collected from pre-diabetic vs. healthy controls according to disease state, age, and sex. 4-week-old Sprague Dawley, male or female rats were fed a high-calorie (HC) or a normal diet for 12 or 24 weeks (wks), representing young vs. old rats, respectively. Female rats fed for 24 wks were further divided into three subcategories: a sham group, one undergoing ovariectomy at wk 12 (OVX), and another receiving estrogen treatment for 12 wks post-surgery (OVX+E2). At weeks 12 or 24, anesthetized rats were instrumented for invasive hemodynamics monitoring via a pressure transducer inserted through the carotid artery. AP signals of length \300 seconds were downsampled to 50Hz, sliced to 1000 data points per bin (\16.7s), and Fourier-transformed. Fourier plots were cropped to remove graph labels/axes and were then fed into the CNN. Our model consisted of 5 convolutional layers, separated by batch normalization, ReLU activation, and max pooling. The outcome of the 5th convolutional layer had 40% of data dropped out and was delivered to a fully connected layer and then to a softmax function. The model utilized cross entropy for a loss function and ADAM for an optimization function. 80% of the data was randomly allocated for training and 10% each for validation and testing. Test accuracy (TA), AUROC, and AUPRC were used to evaluate the model. Our CNN successfully classified rats based on diet (HC vs. control, test accuracy: 97.66%, AUROC: 0.9975, AUPRC: 0.9975), sex (TA: 96.63%, AUROC: 0.9921, AUPRC: 0.9922), surgical outcome (sham vs. OVX vs. OVX+E2, TA: 96.55%, AUROC: 0.9950, AUPRC: 0.9904), and age (16 vs. 28 weeks, TA: 96.65%, AUROC: 0.9802, AUPRC: 0.9757). Upon challenging the model with a 4-way classification task on age and diet (16 wks + control diet, 16 wks + HC, 28 wks + control, and 28 wks + HC), it was able to classify the above with TA: 92.79%, AUROC: 0.9764, and AUPRC: 0.9603. Application of a supervised ML model provides the opportunity of capturing several undefined features of hemodynamic control. Utility of a CNN overcomes the shortcomings associated with the use of other parameters of hemodynamic fluctuations which measure a single characteristic, by probing and compiling additional layers of differences and synthesizing a global appraisal of CV signals.

Finding a link between a hormone and microRNAs (miRNAs) is of great importance since it enables the adjustment of genetic composition or cellular functions without needing gene-level interventions. The dicer-mediated cleavage of precursor miRNAs is an interface link between miRNA and its regulators; any disruption in this process can affect neurogenesis. Besides, the hormonal regulation of miRNAs can occur at the molecular and cellular levels, both directly, through binding to the promoter elements of miRNAs, and indirectly, via regulation of the signaling effects of the post-transcriptional processing proteins. Estrogenic hormones have many roles in regulating miRNAs in the brain. This review discusses miRNAs, their detailed biogenesis, activities, and both the general and estrogen-dependent regulations. Additionally, we highlight the relationship between miR-29, miR-9, and estrogens in the nervous system. Such a relationship could be a possible etiological route for developing various neurodegenerative disorders.