ExhauFS: exhaustive search-based feature selection for classification and survival regression

Spatiotemporal dynamics of cyanobacterial blooms: Integrating machine learning and feature selection techniques with uncrewed aircraft systems and autonomous surface vessel data

Cyanobacterial blooms pose significant threats to aquatic ecosystems and public health due to their ability to release harmful toxins, degrade water quality, disrupt aquatic habitats, and endanger human and animal health through contact or consumption of contaminated water. Monitoring phycocyanin (PC), a pigment unique to cyanobacteria, offers a reliable method for detecting and quantifying these blooms, enabling timely interventions to mitigate their impacts. This study aimed to evaluate ten machine learning algorithms (MLAs) for assessing the spatiotemporal variations of cyanobacterial concentrations over an oyster reef in the Western Mississippi Sound (WMS) using remotely sensed imagery from uncrewed aircraft systems (UAS) and in-situ PC concentrations measured by an autonomous surface vessel (ASV). The study further investigated the influence of river discharge and climatic variables on cyanobacterial concentrations using a time-series of cyanobacteria maps. To derive the most accurate PC retrieval model, a comprehensive set of 85 features was initially generated, including individual spectral bands, band ratios, multiple vegetation indices, and three-band indices. Feature selection was performed using a two-step approach that combined Sequential Backward Floating Selection (SBFS) and Exhaustive Feature Selection (EFS). SBFS was first used to iteratively remove features and optimize model performance, while EFS evaluated all possible combinations of the features identified by SBFS to select the best subset. Among the ten MLAs tested, Extreme Gradient Boosting emerged as the top-performing model, achieving an R2 of 0.835, a root mean square deviation of 0.419 μg/l, an unbiased mean absolute relative difference of 0.176 μg/l, and an average percentage difference of 18.072 % in retrieving PC concentration. The novelty of this study lies in its data-driven approach to identifying the most suitable machine learning algorithm and feature subsets for PC retrieval, thereby enhancing the accuracy and robustness of the developed algorithm. The time-series analysis revealed substantial variations in cyanobacterial concentration in the WMS from 2018 to 2022. The highest average concentration occurred in 2019, coinciding with the introduction of diverted Mississippi River water through the Bonnet Carré Spillway, which triggered an unprecedented cyanobacterial bloom. Furthermore, the average PC concentration was consistently higher during the summer months, likely due to elevated air temperatures and increased sunlight promoting cyanobacterial growth. The methodology developed in this study improves the quantitative monitoring of cyanobacterial blooms using UAS imagery and provides valuable insights for future water quality monitoring initiatives in other regions.

Explainable AI based automated segmentation and multi-stage classification of gastroesophageal reflux using machine learning techniques

Presently, close to two million patients globally succumb to gastrointestinal reflux diseases (GERD). Video endoscopy represents cutting-edge technology in medical imaging, facilitating the diagnosis of various gastrointestinal ailments including stomach ulcers, bleeding, and polyps. However, the abundance of images produced by medical video endoscopy necessitates significant time for doctors to analyze them thoroughly, posing a challenge for manual diagnosis. This challenge has spurred research into computer-aided techniques aimed at diagnosing the plethora of generated images swiftly and accurately. The novelty of the proposed methodology lies in the development of a system tailored for the diagnosis of gastrointestinal diseases. The proposed work used an object detection method called Yolov5 for identifying abnormal region of interest and Deep LabV3+ for segmentation of abnormal regions in GERD. Further, the features are extracted from the segmented image and given as an input to the seven different machine learning classifiers and custom deep neural network model for multi-stage classification of GERD. The DeepLabV3+ attains an excellent segmentation accuracy of 95.2% and an F1 score of 93.3%. The custom dense neural network obtained a classification accuracy of 90.5%. Among the seven different machine learning classifiers, support vector machine (SVM) outperformed with classification accuracy of 87% compared to all other class outperformed combination of object detection, deep learning-based segmentation and machine learning classification enables the timely identification and surveillance of problems associated with GERD for healthcare providers.

Circulating miRNA profiles in COVID-19 patients and meta-analysis: implications for disease progression and prognosis

We compared circulating miRNA profiles of hospitalized COVID-positive patients (n = 104), 27 with acute respiratory distress syndrome (ARDS) and age- and sex-matched healthy controls (n = 18) to identify miRNA signatures associated with COVID and COVID-induced ARDS. Meta-analysis incorporating data from published studies and our data was performed to identify a set of differentially expressed miRNAs in (1) COVID-positive patients versus healthy controls as well as (2) severe (ARDS+) COVID vs moderate COVID. Gene ontology enrichment analysis of the genes these miRNAs interact with identified terms associated with immune response, such as interferon and interleukin signaling, as well as viral genome activities associated with COVID disease and severity. Additionally, we observed downregulation of a cluster of miRNAs located on chromosome 14 (14q32) among all COVID patients. To predict COVID disease and severity, we developed machine learning models that achieved AUC scores between 0.81-0.93 for predicting disease, and between 0.71-0.81 for predicting severity, even across diverse studies with different sample types (plasma versus serum), collection methods, and library preparations. Our findings provide network and top miRNA feature insights into COVID disease progression and contribute to the development of tools for disease prognosis and management.

A novel approach for a joint analysis of isomiR and mRNA expression data reveals features of isomiR targeting in breast cancer

A widely used procedure for selecting significant miRNA-mRNA or isomiR-mRNA pairs out of predicted interactions involves calculating the correlation between expression levels of miRNAs/isomiRs and mRNAs in a series of samples. In this manuscript, we aimed to assess the validity of this procedure by comparing isomiR-mRNA correlation profiles in sets of sequence-based predicted target mRNAs and non-target mRNAs (negative controls). Target prediction was carried out using RNA22 and TargetScan algorithms. Spearman's correlation analysis was conducted using miRNA and mRNA sequencing data of The Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA) project. Luminal A, luminal B, basal-like breast cancer subtypes, and adjacent normal tissue samples were analyzed separately. Using the sets of putative targets and non-targets, we introduced adjusted isomiR targeting activity (ITA)-the number of negatively correlated potential isomiR targets adjusted by the background (estimated using non-target mRNAs). We found that for most isomiRs a significant negative correlation between isomiR-mRNA expression levels appeared more often in a set of predicted targets compared to the non-targets. This trend was detected for both classical seed region binding types (8mer, 7mer-m8, 7mer-A1, 6mer) predicted by TargetScan and the non-classical ones (G:U wobbles and up to one mismatch or unpaired nucleotide within seed sequence) predicted by RNA22. Adjusted ITA distributions were similar for target sites located in 3'-UTRs and coding mRNA sequences, while 5'-UTRs had much lower scores. Finally, we observed strong cancer subtype-specific patterns of isomiR activity, highlighting the differences between breast cancer molecular subtypes and normal tissues. Surprisingly, our target prediction- and correlation-based estimates of isomiR activities were practically non-correlated with the average isomiR expression levels neither in cancerous nor in normal samples.