MICAL is an oxidoreductase that participates in cytoskeleton reorganization via actin disassembly in the presence of NADPH. Although three MICALs (MICAL1, MICAL2 and MICAL3) have been identified in mammals, only the structure of mouse MICAL1 has been reported. Here, the first crystal structure of human MICAL3, which contains the flavin-containing monooxygenase (FMO) and calponin-homology (CH) domains, is reported. MICAL3 has an FAD/NADP-binding Rossmann-fold domain for monooxygenase activity like MICAL1. The FMO and CH domains of both MICAL3 and MICAL1 are highly similar in structure, but superimposition of the two structures shows a different relative position of the CH domain in the asymmetric unit. Based on kinetic analyses, the catalytic efficiency of MICAL3 dramatically increased on adding F-actin only when the CH domain was available. However, this did not occur when two residues, Glu213 and Arg530, were mutated in the FMO and CH domains, respectively. Overall, MICAL3 is structurally highly similar to MICAL1, which suggests that they may adopt the same catalytic mechanism, but the difference in the relative position of the CH domain produces a difference in F-actin substrate specificity.

South Korea’s National Institute of Environmental Research (NIER) operates an algae alert system to monitor water quality at public water supply source sites. Accurate prediction of dominant harmful cyanobacterial genera, such as Aphanizomenon, Anabaena, Oscillatoria, and Microcystis, is crucial for managing water source contamination risks. This study utilized data collected between January 2017 and December 2022 from Juam Lake and Tamjin Lake, which are representative water supply source sites at the Yeongsan River and Seomjin River basins. We performed an exploratory data analysis on the monitored water quality parameters to understand overall fluctuations. Using data from 2017 to 2021 as training data and 2022 data as test data, we compared the dominant algal classification accuracy of 11 statistical machine learning algorithms. The results indicated that the optimal algorithm varied depending on the survey site and evaluation criteria, highlighting the unique environmental characteristics of each site. By predicting dominant algae in advance, stakeholders can better prepare for water source contamination accidents. Our findings demonstrate the applicability of machine learning algorithms as efficient tools for managing water quality in water supply source systems using monitoring data.

To determine the high-priority tributaries that require water quality improvement in the Nakdong River, which is an important drinking water resource for southeastern Korea, data collected at 28 tributaries between 2013 and 2017 were analyzed. To analyze the water quality characteristics of the tributary streams, principal component analysis and factor analysis were performed. COD (chemical oxygen demand), TOC (total organic carbon), TP (total phosphorus), SS (suspended solids), and BOD (biochemical oxygen demand) were classified as the primary factors. In the self-organizing maps analysis using the unsupervised learning neural network model, the first factor showed a highly relevant pattern. To perform the grade classification, 11 parameters were selected. Six parameters are concentrations of the main parameters for the water quality standard assessment in South Korea. We added the pollution load densities for the selected five primary factors. Joochungang showed the highest pollution load density despite its small watershed area. According to the results of the grade classification method, Joochungang, Topyeongcheon, Hwapocheon, Chacheon, Gwangyeocheon, and Geumhogang were selected as tributaries requiring high-priority water quality management measures. From this study, it was concluded that neural network models and grade classification methods could be utilized to identify the high-priority tributaries for more directed and effective water quality management.

No abstract

Ensemble methods often help increase prediction ability in various predictive models by combining multiple weak learners and reducing the variability of the final predictive model. In this work, we demonstrate that ensemble methods also enhance the accuracy of prediction under kernel ridge regression and kernel logistic regression classification. Here we apply bagging and random forests to two kernel-based predictive models; and present the procedure of how bagging and random forests can be embedded in kernel-based predictive models. Our proposals are tested under numerous synthetic and real datasets; subsequently, they are compared with plain kernel-based predictive models and their subsampling approach. Numerical studies demonstrate that ensemble approach outperforms plain kernel-based predictive models.