Suggestion of statistical validation on feature importance of machine learning.

Feature importance methods are widely used in machine learning analysis for medical datasets as both primary and subsidiary tools. These methods aid in selecting biomarkers or markers indicating target diseases, and can provide valuable insight into the mechanism of a disease. However, the simple listing of features with their corresponding importance rank is not sufficient in determining the statistical significance of these features. In this paper, we propose a simple method for evaluating the statistical significance of feature importance values and selecting the optimal number of biomarkers. We demonstrate the application of this method using a public open dataset on heart failure.Clinical Relevance- In order for important indicators to be clinically useful, their statistical significance must be defined. By proposing a simple method for calculating statistical significance, this paper enables clinicians to select a group of biomarkers based on their feature importance in a machine learning model. This approach improves the accuracy and effectiveness of clinical decision-making, leading to more precise diagnosis, treatment, and management of various medical conditions.

Analysis for calibration pre-post difference in BP estimation of Galaxy Watch.

The use of smartwatches has become increasingly common with the release of major products such as the Galaxy Watch by Samsung and the Apple Watch by Apple. The common aim of smartwatches is to target the healthcare market with a wearable, physically-attached device, with blood pressure at the core. As blood pressure is an important biomarker for cardiovascular-related diseases, it is a necessary index to inspect in hospitals when checking an individual's health state. Smartwatches are expected to provide a cuff-less, non-invasive method of estimating blood pressure. However, not many experiments have been conducted on blood pressure datasets obtained from smartwatches. Smartwatches are unique compared to other devices because they require "calibration" to sustain their accuracy.In this paper, we investigate the difference between before and after calibration to better understand the calibration pattern. Not only do we seek to understand the demographic differences in calibration, but we also analyze the possible variables that influence calibration differences. Our results show that hypertensive patients are more prone to high calibration differences, which implies that the calibration period should be adjusted by considering the average blood pressure of users.Clinical Relevance- This paper investigates the possibility for daily BP measurement to be used as clinical data while suggesting proper method to sustain its validity.

Feasibility and measurement stability of smartwatch-based cuffless blood pressure monitoring: A real-world prospective observational study.

Cuffless wearable devices are currently being developed for long-term monitoring of blood pressure (BP) in patients with hypertension and in apparently healthy people. This study evaluated the feasibility and measurement stability of smartwatch-based cuffless BP monitoring in real-world conditions. Users of the first smartwatch-based cuffless BP monitor approved in Korea (Samsung Galaxy Watch) were invited to upload their data from using the device for 4 weeks post calibration. A total of 760 participants (mean age 43.7 ± 11.9, 80.3% men) provided 35,797 BP readings (average monitoring 22 ± 4 days [SD]; average readings 47 ± 42 per participant [median 36]). Each participant obtained 1.5 ± 1.3 readings/day and 19.7% of the participants obtained measurements every day. BP showed considerable variability, mainly depending on the day and time of the measurement. There was a trend towards higher BP levels on Mondays than on other days of the week and on workdays than in weekends. BP readings taken between 00:00 and 04:00 tended to be the lowest, whereas those between 12:00 and 16:00 the highest. The average pre-post calibration error for systolic BP (difference in 7-day BP before and after calibration), was 6.8 ± 5.6 mmHg, and was increased with higher systolic BP levels before calibration. Smartwatch-based cuffless BP monitoring is feasible for out-of-office monitoring in the real-world setting. The stability of BP measurement post calibration and the standardization and optimal time interval for recalibration need further investigation.

Machine learning-based feature selection to search stable microbial biomarkers: application to inflammatory bowel disease.

BACKGROUND: Biomarker discovery exploiting feature importance of machine learning has risen recently in the microbiome landscape with its high predictive performance in several disease states. To have a concrete selection among a high number of features, recursive feature elimination (RFE) has been widely used in the bioinformatics field. However, machine learning-based RFE has factors that decrease the stability of feature selection. In this article, we suggested methods to improve stability while sustaining performance. RESULTS: We exploited the abundance matrices of the gut microbiome (283 taxa at species level and 220 at genus level) to classify between patients with inflammatory bowel disease (IBD) and healthy control (1,569 samples). We found that applying an already published data transformation before RFE improves feature stability significantly. Moreover, we performed an in-depth evaluation of different variants of the data transformation and identify those that demonstrate better improvement in stability while not sacrificing classification performance. To ensure a robust comparison, we evaluated stability using various similarity metrics, distances, the common number of features, and the ability to filter out noise features. We were able to confirm that the mapping by the Bray-Curtis similarity matrix before RFE consistently improves the stability while maintaining good performance. Multilayer perceptron algorithm exhibited the highest performance among 8 different machine learning algorithms when a large number of features (a few hundred) were considered based on the best performance across 100 bootstrapped internal test sets. Conversely, when utilizing only a limited number of biomarkers as a trade-off between optimal performance and method generalizability, the random forest algorithm demonstrated the best performance. Using the optimal pipeline we developed, we identified 14 biomarkers for IBD at the species level and analyzed their roles using Shapley additive explanations. CONCLUSION: Taken together, our work not only showed how to improve biomarker discovery in the metataxonomic field without sacrificing classification performance but also provided useful insights for future comparative studies.

The role of a suburban forest in controlling vertical trace gas and OH reactivity distributions - a case study for the Seoul metropolitan area.

We present trace gas vertical profiles observed by instruments on the NASA DC-8 and at a ground site during the Korea-US air quality study (KORUS) field campaign in May to June 2016. We focus on the region near the Seoul metropolitan area and its surroundings where both anthropogenic and natural emission sources play an important role in local photochemistry. Integrating ground and airborne observations is the major research goal of many atmospheric chemistry field campaigns. Although airborne platforms typically aim to sample from near surface to the free troposphere, it is difficult to fly very close to the surface especially in environments with complex terrain or a populated area. A detailed analysis integrating ground and airborne observations associated with specific concentration footprints indicates that reactive trace gases are quickly oxidized below an altitude of 700 m. The total OH reactivity profile has a rapid decay in the lower part of troposphere from surface to the lowest altitude (700 m) sampled by the NASA DC-8. The decay rate is close to that of very reactive biogenic volatile organic compounds such as monoterpenes. Therefore, we argue that photochemical processes in the bottom of the boundary layer, below the typical altitude of aircraft sampling, should be thoroughly investigated to properly assess ozone and secondary aerosol formation.

The non-classical kinetics and the mutual information of polymer loop formation.

The loop formation of a single polymer chain has served as a model system for various biological and chemical processes. Theories based on the Smoluchowski equation proposed that the rate constant (kloop) of the loop formation would be inversely proportional to viscosity (η), i.e., kloop ∼ η-1. Experiments and simulations showed, however, that kloop showed the fractional viscosity dependence of kloop ∼ η-ß with ß < 1 either in glasses or in low-viscosity solutions. The origin of the fractional viscosity dependence remains elusive and has been attributed to phenomenological aspects. In this paper, we illustrate that the well-known failure of classical kinetics of the loop formation results from the breakdown of the local thermal equilibrium (LTE) approximation and that the mutual information can quantify the breakdown of the LTE successfully.

Fabrication and Evaluation of Cyclic Olefin Copolymer Based Implantable Neural Electrode.

OBJECTIVE: The purpose of this paper is to establish fabrication method of cyclic olefin copolymer(COC)-based neural electrode. METHODS: The fabrication started with preparing COC pellets into COC films by compression molding. Metal layers were deposited on the COC film and attached to a silicon wafer. Laser ablation was used to cut the outer edges and mark alignment keys. The metal layers were patterned using standard photolithography procedures. Finally, the isolated electrodes were laminated. To ensure that the resulting electrode is safe and suitable for long-term implants, in vitro biocompatibility test, impedance evaluation, accelerated soak test, and repeated bend test were conducted. RESULTS: Cytotoxicity test and elution test confirmed the biocompatibility in vitro. The basic performance was not hindered compared to other polymer-based electrodes, and the longevity of the electrode was validated by accelerated soak test. However, repeated bend test revealed that the material might not be suitable for applications where constant bending is required. CONCLUSION: The COC-based neural electrode was successfully fabricated. The material showed several merits such as biocompatibility, thermoplasticity, low water absorption rate, and high transparency, but should be limited to applications where repeated bending is not required. SIGNIFICANCE: Electrical circuits in implantable prosthetic devices must be hermetically encapsulated for a long period of time. Material such as COC with extremely low water absorption rate could have a significant impact on the longevity of these devices.

Spatial Dependence of Non-Gaussian Diffusion of Nanoparticles in Free-Standing Thin Polymer Films.

The addition of nanoparticles (NPs) to a free-standing polymer film affects the properties of the film such as viscosity and glass transition temperature. Recent experiments, for example, showed that the glass transition temperature of thin polymer films was dependent on how NPs were distributed within the polymer films. However, the spatial arrangement of NPs in free-standing polymer films and its effect on the diffusion of NPs and polymers remain elusive at a molecular level. In this study, we employ generic coarse-grained models for polymers and NPs and perform extensive molecular dynamics simulations to investigate the diffusion of polymers and NPs in free-standing thin polymer films. We find that small NPs are likely to stay at the interfacial region of the polymer film, while large NPs tend to stay at the center of the film. On the other hand, as the interaction between a NP and a monomer becomes more attractive, the NP is more likely to be placed at the film center. The diffusion of monomers slows down slightly as more NPs are added to the film. Interestingly, the NP diffusion is dependent strongly on the spatial arrangement of the NPs: NPs at the interfacial region diffuse faster and undergo more non-Gaussian diffusion than NPs at the film center, which implies that the interfacial region would be more mobile and dynamically heterogeneous than the film center. We also find that the mechanism for non-Gaussian diffusion of NPs at the film center differs from that at the interfacial region and that the NP diffusion would reflect the local viscosity of the polymer films.

Design of Microscope Optics for the Acquisition of Multiple Wavelength Band Image.

We design combinations of filters that lead to make outputs of multiple wavelength bands simultaneously. These bands are either 1) Infrared, RGB, Ultraviolet-3ranges 2) IR, Red, Green, Blue, UV-5ranges 3) IR+ Red, Green, Blue+UV-3ranges. Each is mostly devised by the arrangement of normal cold mirror, UV cold mirror, Philips(RGB) prism, UV filter and color filter. For the selection 1), UV cold mirror reflects UV and transmit VIS + IR. VIS and IR are sorted by normal cold mirror, which passes IR and reflects VIS. Each sorted light goes to each sensor to capture the images. The sorting of VIS and UV also can be done by UV filter. For the selection 2), have a same process in 1) but make VIS pass Philips(RGB) prism which sorts VIS into R, G, B. For the selection 3), the light passes green color filter first, Green color is only transmitted and the others are reflected. And then reflected light goes through UV cold mirror so that UV and Blue (wavelength <; 500nm) are reflected. And the others(wavelength>550nm, but green color zone was already sorted roughly wavelength>600nm) are transmitted, which is Red and IR. This sorting can be done by another design. After sorting all ranges by the selection 2), we can arrange UV goes a same pass with blue, and IR with red. We need band pass filters (in here, color filters) to revise the passes.