Fit To Data Sets Research Articles

FLIGHTED: Inferring Fitness Landscapes from Noisy High-Throughput Experimental Data.

Machine learning (ML) for protein design requires large protein fitness datasets generated by high-throughput experiments for training, fine-tuning, and benchmarking models. However, most models do not account for experimental noise inherent in these datasets, harming model performance and changing model rankings in benchmarking studies. Here we develop FLIGHTED, a Bayesian method of accounting for uncertainty by generating probabilistic fitness landscapes from noisy high-throughput experiments. We demonstrate how FLIGHTED can improve model performance on two categories of experiments: single-step selection assays, such as phage display and SELEX, and a novel high-throughput assay called DHARMA that ties activity to base editing. We then compare the performance of standard machine-learning models on fitness landscapes generated with and without FLIGHTED. Accounting for noise significantly improves model performance, especially of CNN architectures, and changes relative rankings on numerous common benchmarks. Based on our new benchmarking with FLIGHTED, data size, not model scale, currently appears to be limiting the performance of protein fitness models, and the choice of top model architecture matters more than the protein language model embedding. Collectively, our results indicate that FLIGHTED can be applied to any high-throughput assay and any machine learning model, making it straightforward for protein designers to account for experimental noise when modeling protein fitness.

Transmuted cosine Topp-Leone G family of distributions: properties and applications

In contemporary data analysis, the need for flexible and adaptable probabilistic models capable of capturing complex dependencies and tail behaviors in real-life datasets has become increasingly apparent. In light of this demand, A new trigonometric generalized family of distribution, the "Transmuted Cosine Topp-Leone G Family" is proposed in this study. Established on the foundations of, this family combines the adaptability of the Topp-Leone distribution with the periodicity of the cosine function and transmutation theory to produce a flexible framework that may be used to represent a wide range of real-life phenomena. We derive some of the statistical properties of the introduced family such as; survival and hazard functions, and moment and moment-generating functions. Moreover, the model parameters are estimated using the Maximum Likelihood method, and a Monte Carlo simulation was performed to ascertain the behavior and the consistency of the estimates. We investigate the shape of the distribution parameters which results in distributions with left-skew, right-skew, increasing, and decreasing shapes. Lastly, through empirical demonstrations, we showcase the efficacy of the family TrCTLG models in fitting lifetime datasets.

A Transferable Deep Learning Framework for Improving the Accuracy of Internet of Things Intrusion Detection

As the size of the IoT solutions and services market proliferates, industrial fields utilizing IoT devices are also diversifying. However, the proliferation of IoT devices, often intertwined with users’ personal information and privacy, has led to a continuous surge in attacks targeting these devices. However, conventional network-level intrusion detection systems with pre-defined rulesets are gradually losing their efficacy due to the heterogeneous environments of IoT ecosystems. To address such security concerns, researchers have utilized ML-based network-level intrusion detection techniques. Specifically, transfer learning has been dedicated to identifying unforeseen malicious traffic in IoT environments based on knowledge distillation from the rich source domain data sets. Nevertheless, since most IoT devices operate in heterogeneous but small-scale environments, such as home networks, selecting adequate source domains for learning proves challenging. This paper introduces a framework designed to tackle this issue. In instances where assessing an adequate data set through pre-learning using transfer learning is non-trivial, our proposed framework advocates the selection of a data set as the source domain for transfer learning. This selection process aims to determine the appropriateness of implementing transfer learning, offering the best practice in such scenarios. Our evaluation demonstrates that the proposed framework successfully chooses a fitting source domain data set, delivering the highest accuracy.

Machine learning analysis of RB-TnSeq fitness data predicts functional gene modules in Pseudomonas putida KT2440.

This study demonstrates a rapid, automated approach for elucidating functional modules within complex genetic networks. While Pseudomonas putida randomly barcoded transposon insertion sequencing data were used as a proof of concept, this approach is applicable to any organism with existing functional genomics data sets and may serve as a useful tool for many valuable applications, such as guiding metabolic engineering efforts in other microbes or understanding functional relationships between virulence-associated genes in pathogenic microbes. Furthermore, this work demonstrates that comparison of data obtained from independent component analysis of transcriptomics and gene fitness datasets can elucidate regulatory-functional relationships between genes, which may have utility in a variety of applications, such as metabolic modeling, strain engineering, or identification of antimicrobial drug targets.

A Comparative Study of Principled rPPG-Based Pulse Rate Tracking Algorithms for Fitness Activities.

Performance improvements obtained by recent principled approaches for pulse rate(PR) estimation have typically been achieved by adding or modifying certain modules within a reconfigurable system. Yet, evaluations are usually performed only at the system level. To better understand each module's contribution and facilitate future research in explainable learning and artificial intelligence for physiological monitoring, this paper conducts a comparative study of video-based, principled PR tracking algorithms, with a particular focus on challenging fitness scenarios. A review of the progress achieved over the last decade and a half in this field is utilized to construct the major processing modules of a reconfigurable remote pulse rate sensing system. Experiments are conducted on two challenging datasets-an internal collection of 25 videos of two Asian males exercising on stationary-bike, elliptical, and treadmill machines and 34 videos from a public ECG fitness database of 14 men and 3 women exercising on elliptical and stationary-bike machines. The signal-to-noise ratio (SNR), Pearson's correlation coefficient, error count ratio, error rate, and root mean squared error are used for performance evaluation. The top-performing configuration produces respective values of -0.8dB, 0.86, 9%, 1.7%, and 3.3beats per minute(bpm) for the internal dataset and 1.3dB, 0.77, 28.6%, 6.0%, and 8.1bpm for the ECG Fitness dataset, achieving significant improvements over alternative configurations. Our results indicate a synergistic effect between pulse color mapping and adaptive motion filtering, as well as the importance of a robust frequency tracking algorithm for PR estimation in low SNR settings.

SATLAS2: An update to the package for analysis of counting data

SATLAS2 is a Python library that enables the user to fit counting data from laser spectroscopy experiments, in particular those that measure atomic hyperfine structures. In this analysis, the user can choose how the uncertainties are treated and can also opt to generate a random walk in order to present a fuller picture of the parameter space. The major upgrade compared to the previous version of SATLAS [1] is the different architecture of the codebase, which enabled a performance boost, with speed-up factors ranging from 20 to 300 times for various use cases. For backward compatibility, a translation layer between the two architectures is available, implementing only the core functionality of SATLAS. New version program summaryProgram Title: SATLAS2CPC Library link to program files:https://doi.org/10.17632/3hr8f5nkhb.2Developer's repository link:https://github.com/IKS-nm/satlas2Licensing provisions: MITProgramming language: PythonJournal reference of previous version: Computer Physics Communications, Volume 222, 2018, Pages 286-294, ISSN 0010-4655, https://doi.org/10.1016/j.cpc.2017.09.012Does the new version supersede the previous version?: YesReasons for the new version: Improved and more stable performanceSummary of revisions: The architecture of the SATLAS package has been completely changed. Instead of operating on Models, using SumModel and LinkedModel for various analysis options, SATLAS2 works with a central Fitter object to which Sources get assigned, which themselves get assigned Models. This structure improves the performance by streamlining the code and benefits from using a pass by reference approach for changing the parameters, rather than a pass by value. Most of the functionality is maintained, with a small interface for most basic things available so SATLAS2 can be used as a drop-in replacement.Nature of problem: The analysis of specifically counting data has some special considerations compared to more common datasets with Gaussian-distributed uncertainties. Application of Bayesian inference through random walk exploration is useful for more accurate exploration of parameter space. The fitting of multiple models with shared parameters can be desirable for either fitting multiple datasets with the same parameters or imposing additional restrictions on the parameters.Solution method: SATLAS2 implements the correct statistical costs for fitting of both counting data and Gaussian distributed uncertainties, and allows an easy implementation of custom cost functions.. Through the implementation of multiple Sources and Models, a natural extension of the fitting is made, so multiple models can be both summed together to fit to the same data and multiple fits can be linked together through parameter linking, using the underlying LMFIT [2] library. The ability to fit simultaneously increases the possibilities for extensive statistical analyses, and the interface through LMFIT with the emcee [3] library enables Bayesian exploration of the parameter space.

Machine learning prediction and interpretation of the impact of microplastics on soil properties

The annual microplastic (MP) release into soils is 4–23 times higher than that into oceans, significantly impacting soil quality. However, the mechanisms underlying how MPs impact soil properties remain largely unknown. Soil-MP interactions are complex because of soil heterogeneity and varying MP properties. This lack of understanding was exacerbated by the diverse experimental conditions and soil types used in this study. Predicting changes in soil properties in the presence of MPs is challenging, laborious, and time-consuming. To address these issues, machine learning was applied to fit datasets from peer-reviewed publications to predict and interpret how MPs influence soil properties, including pH, dissolved organic carbon (DOC), total P, NO3−-N, NH4+-N, and acid phosphatase enzyme activity (acid P). Among the developed models, the gradient boost regression (GBR) model showed the highest R2 (0.86–0.99) compared to the decision tree and random forest models. The GBR model interpretation showed that MP properties contributed more than 50% to altering the acid P and NO3−-N concentrations in soils, whereas they had a negligible impact on total P and 10–20% impact on soil pH, DOC, and NH4+-N. Specifically, the size of MPs was the dominant factor influencing acid P (89.3%), pH (71.6%), and DOC (44.5%) in soils. NO3−-N was mainly affected by the MP type (52.0%). The NH4+-N was mainly affected by the MP dose (46.8%). The quantitative insights into the impact of MPs on soil properties of this study could aid in understanding the roles of MPs in soil systems.

Predict The Caloric Expenditure and Pose Estimation Through the Assistance of The Virtual Gym Coach

This research paper introduces a novel virtual gym assistant leveraging Google's Mediapipe library, designed for diverse multimodal machine learning and deep learning pipelines. The system offers real-time guidance by analyzing user movements during specific exercises using posture estimation algorithms. Developed with Mediapipe's deep algorithms and pose estimation module, the system captures user movements through the identification of body landmarks, facilitating rep counting for each exercise. Angles and landmarks are then processed and transmitted to various machine learning models, enabling the classification of correct postures and reps. Through machine learning algorithms, the system evaluates the accuracy of user-performed reps. Utilizing diverse fitness datasets and custom data, the results demonstrate that the proposed system delivers precise and personalized feedback, enhancing exercise performance and minimizing the risk of injury. Additionally, the system calculates caloric expenditure, providing comprehensive support for users through the Virtual Gym Coach.

Fast and accurate prediction of material properties with three-body tight-binding model for the periodic table.

Parametrized tight-binding models fit to first-principles calculations can provide an efficient and accurate quantum mechanical method for predicting properties of molecules and solids. However, well-tested parameter sets are generally only available for a limited number of atom combinations, making routine use of this method difficult. Furthermore, many previous models consider only simple two-body interactions, which limits accuracy. To tackle these challenges, we develop a density functional theory database of nearly 1 000 000 materials, which we use to fit a universal set of tight-binding parameters for 65 elements and their binary combinations. We include both two-body and three-body effective interaction terms in our model, plus self-consistent charge transfer, enabling our model to work for metallic, covalent, and ionic bonds with the same parameter set. To ensure predictive power, we adopt a learning framework where we repeatedly test the model on new low-energy crystal structures and then add them to the fitting data set, iterating until predictions improve. We distribute the materials database and tools developed in this paper publicly.

Privacy-Preserving Personalized Fitness Recommender System P 3 FitRec : A Multi-level Deep Learning Approach

Recommender systems have been successfully used in many domains with the help of machine learning algorithms. However, such applications tend to use multi-dimensional user data, which has raised widespread concerns about the breach of users’ privacy. Meanwhile, wearable technologies have enabled users to collect fitness-related data through embedded sensors to monitor their conditions or achieve personalized fitness goals. In this article, we propose a novel privacy-aware personalized fitness recommender system. We introduce a multi-level deep learning framework that learns important features from a large-scale real fitness dataset that is collected from wearable Internet of Things (IoT) devices to derive intelligent fitness recommendations. Unlike most existing approaches, our approach achieves personalization by inferring the fitness characteristics of users from sensory data, minimizing the need for explicitly collecting user identity or biometric information, such as name, age, height, and weight. Our proposed models and algorithms predict (a) personalized exercise distance recommendations to help users to achieve target calories, (b) personalized speed sequence recommendations to adjust exercise speed given the nature of the exercise and the chosen route, and (c) personalized heart rate sequence to guide the user of the potential health status for future exercises. Our experimental evaluation on a real-world Fitbit dataset demonstrated high accuracy in predicting exercise distance, speed sequence, and heart rate sequence compared with similar studies. 1 Furthermore, our approach is novel compared with existing studies, as it does not require collecting and using users’ sensitive information. Thus, it preserves the users’ privacy.

Theoretical Structure and Applications of a Newly Enhanced Gumbel Type II Model

Statistical models are vital in data analysis, and researchers are always on the search for potential or the latest statistical models to fit data sets in a variety of domains. To create an improved statistical model, we used a T-X transformation and the Gumbel Type-II model in this investigation. The research examined a simulation evaluation to assess the efficacy of the parameters. To show the application of the T-X approach for producing new distributions titled the new and improved Gumbel Type-II (NIGT-II) distribution, two actual data sets were used. The data sets reveal that the NIGT-II distribution sounds nicer than the Gumbel Type-II distribution.

Large-scale genetic characterization of the model sulfate-reducing bacterium, Desulfovibrio vulgaris Hildenborough.

Sulfate-reducing bacteria (SRB) are obligate anaerobes that can couple their growth to the reduction of sulfate. Despite the importance of SRB to global nutrient cycles and their damage to the petroleum industry, our molecular understanding of their physiology remains limited. To systematically provide new insights into SRB biology, we generated a randomly barcoded transposon mutant library in the model SRB Desulfovibrio vulgaris Hildenborough (DvH) and used this genome-wide resource to assay the importance of its genes under a range of metabolic and stress conditions. In addition to defining the essential gene set of DvH, we identified a conditional phenotype for 1,137 non-essential genes. Through examination of these conditional phenotypes, we were able to make a number of novel insights into our molecular understanding of DvH, including how this bacterium synthesizes vitamins. For example, we identified DVU0867 as an atypical L-aspartate decarboxylase required for the synthesis of pantothenic acid, provided the first experimental evidence that biotin synthesis in DvH occurs via a specialized acyl carrier protein and without methyl esters, and demonstrated that the uncharacterized dehydrogenase DVU0826:DVU0827 is necessary for the synthesis of pyridoxal phosphate. In addition, we used the mutant fitness data to identify genes involved in the assimilation of diverse nitrogen sources and gained insights into the mechanism of inhibition of chlorate and molybdate. Our large-scale fitness dataset and RB-TnSeq mutant library are community-wide resources that can be used to generate further testable hypotheses into the gene functions of this environmentally and industrially important group of bacteria.

Globally optimal univariate spline approximations

We revisit the problem of computing optimal spline approximations for univariate least-squares splines from a combinatorial optimization perspective. In contrast to most approaches from the literature we aim at globally optimal coefficients as well as a globally optimal placement of a fixed number of knots for a discrete variant of this problem. To achieve this, two different possibilities are developed. The first approach that we present is the formulation of the problem as a mixed-integer quadratically constrained problem, which can be solved using commercial optimization solvers. The second method that we propose is a branch-and-bound algorithm tailored specifically to the combinatorial formulation. We compare our algorithmic approaches empirically on both, real and synthetic curve fitting data sets from the literature. The numerical experiments show that our approach to tackle the least-squares spline approximation problem with free knots is able to compute solutions to problems of realistic sizes within reasonable computing times.

Optimizing height measurement for the long-term forest experiments in Sweden

Information on tree height is useful for volume estimation and site productivity assessment and as such, remains one of the most important variables often measured in forest inventories. Measuring a sufficient number of sample trees requires considerable sampling effort and cost. In this study, we developed height functions for optimizing tree height measurement in the Swedish long-term forest experiments (LTFEs). Two large datasets from the LTFE databases: fitting data (from thinning, fertilisation and mixed species experiments) and validation data (tree species and spacing experiments) collected over several decades were used. The fitting and validation data comprise 133,788 and 68,440 observations, respectively, each covering a large range of growth and environmental conditions across Sweden. A multilevel nonlinear mixed-effects modelling approach was used to build the generalised height functions for Scots pine, Norway spruce, birch (Silver and Downy birch united), other conifers and other broadleaves, considering variations in heights and other stand characteristics at sample plot-level and revision-level. The response calibration of the functions was first carried out with all measured heights of the validation data, and second, using heights of one to six sample trees obtained from different tree selection strategies (diameter extremes, largest diameters, and smallest diameters). The mixed-effects height functions explained most of the height variations in the fitting dataset (pseudo R2: 0.938 – 0.970; RMSE: 0.957 – 1.363 m) without any residual trends. The validation showed that the functions accounted for 95 – 98 % of the height variation in the validation dataset, with RMSE ranging between 0.770 and 1.040 m, confirming the functions’ high accuracy. We recommend the measurement of four sample tree heights based on diameter extremes as the ideal threshold for response calibration. These functions and the suggested sampling technique would reduce sampling effort and inventory cost of height measurements for subsequent inventories of the LTFEs.

Sparse Data Traffic Speed Prediction on a Road Network With Varying Speed Levels

Most works on graph neural networks (GNNs) for traffic speed prediction assume near-complete data and little variance of base speed levels. However, both assumptions do not necessarily hold true for network-wide probe vehicle data (PVD). Therefore, we applied two state-of-the-art GNNs to sparse PVD from a road network with highly varying speed levels and to dense motorway data for comparison. We introduce two methods to adapt preexisting GNNs for improved prediction performance: normalization of speed values with respect to the base speed levels of different roads led to significant improvements of prediction performance on both datasets. Using the number of observations supporting each speed value as an additional input feature can improve prediction performance. Furthermore, we identified characteristics of data and models encouraging the use of either method. As no fitting dataset to evaluate these approaches was found, a novel dataset derived from PVD is introduced. It features sparse speed values and underlying numbers of observations for a road network with varying speed levels.

Use of time dependent data in Bayesian global 21-cm foreground and signal modelling

ABSTRACT Global 21-cm cosmology aims to investigate the cosmic dawn and epoch of reionization by measuring the sky averaged H i absorption signal, which requires, accurate modelling of, or correction for, the bright radio foregrounds and distortions arising from chromaticity of the antenna beam. We investigate the effect of improving foreground modelling by fitting data sets from many observation times simultaneously in a single Bayesian analysis, fitting for the same parameter set by performing these fits on simulated data. We find that for a hexagonal dipole antenna, this simultaneous fitting produces a significant improvement in the accuracy of the recovered 21-cm signal, relative to fitting a time average of the data. Furthermore, the recovered models of the foreground are also seen to become more accurate by up to a factor of ∼2–3 relative to time averaged fitting. For a less chromatic log spiral antenna, no significant improvement in signal recovery was found by this process. However, the modelling of the foregrounds was still significantly improved. We also investigate extending this technique to fit multiple data sets from different antennas simultaneously for the same parameters. This is also found to improve both 21-cm signal and foreground modelling, to a higher degree than fitting data set from multiple times from the same antenna.

CFPNet: A Denoising Network for Complex Frequency Band Signal Processing

The recent development of deep learning has brought breakthroughs in image denoising. However, the recovery of image detail, especially high-frequency weak information, still needs to be improved. Firstly, the noise mainly concentrates on the high-frequency signal, and the high-frequency signal is easy to be disturbed, which makes it difficult to recover; Secondly, in the process of image denoising with deep learning, feature extraction of model is used to smooth the noise for image restoration, resulting in a poor recovery effect of high-frequency signal. To solve the above problems and improve the overall image denoising performance, we propose a denoising network for complex frequency band signal processing (CFPNet), which contains three insights: 1) the image input node uses a cosine transform to segment the image noise frequency and divides different image features into signals in different frequency bands for targeted noise reduction; 2) targeted noise reduction is carried out for different frequency band signals via a fine-grained scheme; 3) different frequency band signals are fused and high-frequency signals are enhanced to improve the recovery of detailed signals. The experimental results show that the proposed CFPNet can achieve state-of-the-art performance on both real-world datasets and Gaussian noise fitting datasets.

Exponentiated Complementary Kumaraswamy-G Power Series Family of Distributions with Application to Lifetime Data

This paper proposed the Exponentiated Complementary Kumaraswamy - G Power (ECK-GP) Series family of distributions which is an extension of the Complementary Kumaraswamy-G Power Series family of distributions. The expressions for its densities, basic statistical properties and parameters estimation using the method of Maximum Likelihood were derived and established. An application of the Exponentiated Complementary Kumaraswamy Exponential Poisson (ECKEP) distribution to a real lifetime dataset clearly reveals its suitability and flexibility in fitting real life dataset.

Laplace transformed properties of the extended power-Gompertz model: Simulation and applications

In this paper, a new four-parameter lifetime model tilted the exponentiated power Gompertz (EXPGO) model is proposed. The explicit expressions of the density, distribution and quantile functions of the EXPGO are provided. Its density function can be expressed as a linear mixture of the generalized Gompertz densities, and can be unimodal and left skewed. The failure rate of the new model can be concave decreasing and reversed J-shaped. Likewise, the structural properties of the new model like the moments, moments generating function, probability weighted moments and entropies expressed in Laplace transform are provided. The maximum likelihood estimator for the new model is discussed and the performance of the estimator evaluated using Monte Carlo simulations. Three real-data applications indicate that the EXPGO model provides a better fit and can be utilized in fitting skewed datasets with unimodal features and concave decreasing failure rates.

Symbolic regression assisted design of perovskite-type OER electrocatalysts for oxygen evolution reaction (OER) in alkaline electrolyte

Perovskite oxides are promising electrocatalysts for oxygen evolution reaction (OER) in alkaline electrolytes. With high activity and low cost, they are being pursued by researchers worldwide. However, due to the compositional and structural diversity of perovskite-type oxides, the conventional method based on experimental trials on thousands of elemental combinations is inefficient and time-consuming, which hampers the development of this type of electrocatalyst. Here we employed a symbolic regression (SR) method under a machine learning framework to build mathematical formulae that best fit datasets based on 105 experimentally studied perovskite-type electrocatalysts and identified a simple parameter as the OER activity descriptor. Results show that B-site cations with a large number of d electrons and a small radius, such as Ni4+, Co4+, and Fe4+, are beneficial to high OER activity in perovskite oxides, which provide a guide for materials design.