Glaucoma optic neuropathy (GON) is a glaucoma complication that has become the second most common cause of blindness in the world. This condition occurs due to an uncontrolled increase in intraocular pressure. Current GON risk monitoring still requires direct contact with the eyes (invasive) and is only available in healthcare centers, so patients are prone to drop out of therapy. A GON risk assessment system that fulfills eye care device standards is needed. Glaucoma Assist (Glassist), a monitoring assistant system based on a biometry model and machine learning integration, was developed as a GON risk monitoring tool. It is inspired by a hollow sphere diameter that will increase with its pressure. It is also directly proportional to the biometric model of axial length changes. This concept is integrated with object detection machine learning and displays the GON risk interpretation on a website. Glassist and iCare tonometer showed a matched result with the error of 0 ± 0.8 mmHg, which fulfills the allowable tolerance by ISO 8612:2009. Glassist’s features are easy to use, showing an overall success rate of 86%. The system usability scale test indicates that Glassist can function well and is accepted by the users with a final score of 82%. Glassist has fulfilled eye care device standards and is highly potent for commercialization.
Protein structure prediction (PSP) is a fundamental challenge in computational biology, essential for understanding molecular mechanisms and accelerating drug discovery. This systematic review, conducted under the PRISMA guidelines, presents the application of machine learning (ML) methods for PSP, with a focus on deep learning models and hybrid approaches from 2014 to 2025. A comprehensive search across major databases, retrieved 1,939 studies, of which 43 met the inclusion criteria for full-text analysis. The studies reviewed employed state-of-the-art ML techniques such as Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), Random Forests (RF), and ensemble models. Advanced methods like AlphaFold and RoseTTAFold were highlighted for their accuracy in tertiary structure prediction, with TM-scores surpassing 0.7. Other models like ThreaderAI and DeepMSA2 demonstrated significant advancements in template-based modeling and secondary structure prediction. The analysis identified common challenges, including dataset biases primarily linked to well-characterized proteins from the Protein Data Bank (PDB), limited performance in predicting intrinsically disordered proteins (IDPs), and the lack of interpretability in deep learning models. Few studies integrated Explainable AI (XAI) techniques to enhance model transparency, indicating an area for future development. In conclusion, this systematic review provides insights into current ML-driven methodologies for PSP, outlines key challenges, and suggests the need for improved dataset diversity, explainable models, and hybrid approaches to bridge the gap between prediction and biological interpretation.
This study aims to explore the regional utilization of organic household waste composting and its resource recycling. Taking Zhoukou city in Henan Province, China, as the case study, the research focuses on two districts, one county-level city, and seven counties within the city’s administrative boundaries. Using methods such as literature review, field investigation, data collection, and quantitative analysis, the study quantitatively examines the demand-supply balance of organic waste conversion into compost for greening in each research region. In 2021, the total potential demand for greening compost in Zhoukou city was approximately 247.28 kt/a, while the total potential supply of organic waste compost was about 426.54 kt/a, resulting in a demand-supply ratio of 0.58. The demand was found to be lower than the supply, indicating that not all organic household waste compost can be utilized for greening purposes across the city, and there are significant deviations in the demand-supply ratio between different regions. Furthermore, assuming actual supply rates of 85% and 70% for the organic waste compost, the demand-supply balance was analyzed under these conditions. The results show that when the actual supply rate is 85% or 70%, the demand-supply balance improves, with a more pronounced increase at a 70% supply rate.
This work develops a Bayesian uncertainty quantification framework for personalized modeling of epileptiform seizure dynamics using a high order discontinuous Galerkin discretization of the monodomain reaction diffusion equation coupled with the Barreto-Cressman conductance based ionic model. The forward model simulates transmembrane potential propagation in anisotropic neural tissue, incorporating complex brain geometries and ionic kinetics. To calibrate patient-specific biophysical parameters and capture modeling uncertainty, the framework embeds the forward system into a Bayesian inference pipeline using the Metropolis-Hastings algorithm known as Markov chain Monte Carlo sampling. Parameters such as ion channel conductance and gating time constants are inferred from noisy synthetic voltage data, with posterior distributions evaluated through convergence diagnostics, posterior predictive checks, and estimation error analysis. Results demonstrate accurate recovery of dominant conductance, partial identifiability of gating kinetics, and robust predictive performance in modeling transient seizure activity. The proposed framework enables uncertainty-aware, patient-specific calibration of electrophysiological models and supports future applications in seizure forecasting and neurostimulation planning.
