Unbiased Model Research Articles

Genetic architecture and genomic prediction of plant height-related traits in chrysanthemum.

Plant height (PH) is a crucial trait determining plant architecture in chrysanthemum. To better understand the genetic basis of PH, we investigated the variations of PH, internode number (IN), internode length (IL), and stem diameter (SD) in a panel of 200 cut chrysanthemum accessions. Based on 330 710 high-quality SNPs generated by genotyping by sequencing, a total of 42 associations were identified via a genome-wide association study (GWAS), and 16 genomic regions covering 2.57Mb of the whole genome were detected through selective sweep analysis. In addition, two SNPs, Chr1_339370594 and Chr18_230810045, respectively associated with PH and SD, overlapped with the selective sweep regions from FST and π ratios. Moreover, candidate genes involved in hormones, growth, transcriptional regulation, and metabolic processes were highlighted based on the annotation of homologous genes in Arabidopsis and transcriptomes in chrysanthemum. Finally, genomic selection for four PH-related traits was performed using a ridge regression best linear unbiased predictor model (rrBLUP) and six marker sets. The marker set constituting the top 1000 most significant SNPs identified via GWAS showed higher predictabilities for the four PH-related traits, ranging from 0.94 to 0.97. These findings improve our knowledge of the genetic basis of PH and provide valuable markers that could be applied in chrysanthemum genomic selection breeding programs.

122 Comparison of Algorithms for Approximation of Accuracies for Single-Step Genomic Best Linear Unbiased Predictor Models

Abstract The theoretical accuracy of breeding values can be calculated based on the prediction error variances obtained from the diagonal of the inverse of the left-hand side of the mixed model equations. However, inverting the coefficient matrix is not computationally feasible for an extensive system of equations, especially if genomic information is available. Thus, different algorithms to approximate accuracies have been proposed. We aimed to compare the efficiency of two algorithms implemented in the accf90GS and accf90GS2 software from the BLUPF90 family. The accf90GS algorithm approximates accuracies based on the diagonal of the genomic relationship matrix (G). In turn, the accf90GS2 uses block sparse inversion of G-1. The data for this study were provided by the American Angus Association (Saint Joseph, MO) and included phenotypes of three growth traits and eleven carcass traits, 1,235,930 genotyped animals, and pedigree ranging from 4,315,913 to 12,492,581 animals. Three analyses were performed: 1) comparison of the approximated accuracies between algorithms; 2) comparison between approximated and exact accuracies; and 3) evaluation of the impact of adding genotyped animals, with and without phenotypes, in the approximated accuracies. The results were obtained for genotyped animals (group 1) and genotyped animals without phenotypes (group 2). The complete datasets under multi-trait models with ssGBLUP and the algorithm for proven and young were used for analysis 1. A subset of data under single-trait models was used for analyses 2 and 3, so it was possible to calculate the exact accuracies. In analysis 1, the correlations of groups 1 and 2 ranged from 0.55 to 0.74 and 0.38 to 0.65, respectively. The growth traits presented the lowest correlations. In addition, the slope was under 0.75 for both groups. These results suggest that the approximated accuracies are considerably different, especially for growth traits and genotyped animals without phenotype. The accf90GS2 algorithm presented the highest correlations with exact accuracies, which indicates that it is more efficient than the accg90GS algorithm. The intercept was almost zero for both approximation methods, whereas the slope ranged from 0.937 to 1.234 and 0.819 to 0.949 for accf90GS and accf90GS2 accuracies, respectively. This suggests a minor overestimation and underestimation by the accf90GS2 and accf90GS algorithms, respectively. The increase in mean accuracy due to adding genotyped animals with phenotypes was higher for accf90GS2. According to the results, the algorithm implemented in accf90GS2 is recommended for genetic evaluations because of its efficiency and precision.

PSV-6 The Impacts of Heat Stress on Litter Traits of Landrace Population in Taiwan

Abstract Global warming becomes a serious issue in the swine industries of most countries. Sow reproduction suffers the brunt of heat stress problems in tropical and subtropical zones, such as in Taiwan. Litter traits, total number born (TNB) and number born alive (NBA), are essential for sow stability and farm economics. However, these traits are influenced by management, facilities, and environmental conditions. Adjustments to management practices and facilities must be considered environmental conditions, including temperature and relative humidity, particularly in Taiwan under oceanic climates. Such specific climate with high temperature and relative humidity caused heat stress for sows around Taiwan over six months in a year. Therefore, the impact of heat stress at different stages of pregnancy in sows had been analyzed using the best linear unbiased prediction model with heterogeneous residuals. A total of 11,074 records for TNB and NBA from 3,080 Landrace sows were collected from two farms between 2008 and 2021. Simultaneously, weather data were obtained from those weather stations closest to the farms. Sows from mating to farrowing, about four months, had been classified into cool (C) or hot (H) climate condition in each month based on temperature-humidity index equals 70. Hence, there were eight combinations (CCCC, CCCH, CCHH, CHHH, HHHH, HHHC, HHCC, HCCC) of the climate effect in this study. The results of climate effect showed that there was not significant difference among the groups of CCCC, CCCH, CCHH, and CHHH in TNB and NBA, while no significant difference (P > 0.05) was found among the groups of HHHH, HHHC, HHCC, and HCCC in both traits. However, the climate effects of CCCH, CCHH, and CHHH were greater (P < 0.05) than those of HHHC, HHCC, and HCCC in TNB. The climate effects of CCCH and CCHH were greater than those of HHHH, HHHC, HHCC, and HCCC in NBA (P < 0.05). These fluctuations generally indicated when sows experience heat stress on mating and early stage of pregnancy (HHHH, HHHC, HHCC, HCCC), could decrease litter sizes. In the heterogeneous residual variances, the groups of CCCC, CCCH, CCHH, and CHHH were less than the groups of HHHH, HHHC, and HHCC in TNB, significantly, while the groups of CCCC, CCCH, and CHHH were less than the groups of HHHH, HHHC, and HHCC in NBA (P < 0.05). The greatest estimated residual variances were shown in HHCC in both TNB and NBA. These changing of trends indicated high heterogeneous residual variance was discovered when heat stress happened during mating and early pregnancy of sows (HHHH, HHHC, HHCC). In conclusion, heat stress at mating and early stage of sow pregnancy decreased the average and increased the residual variances of TNB and NBA.

Probabilistic burst pressure prediction model for pipelines with single crack-like defect

While several models have been developed to predict the failure pressure of pipelines with corrosion defects, relatively few models are available for pipelines with cracking defects. The goal of this study is to develop an unbiased and accurate probabilistic burst failure pressure model for thin-walled pipelines containing a single, longitudinally-oriented, crack-like defect. A comprehensive database is first established, consisting of experimental and numerical data collected from literature as well as numerical data obtained from the extended finite element method (XFEM) conducted in this study. The failure pressure prediction model is developed by adding a correction factor to the modified Ln-Secant method, a semi-empirical model previously developed. The correction factor is modeled using a multivariate linear regression that considers the pipe properties and defect characteristics as independent variables. The proposed model delivers unbiased and less variable failure pressure predictions. A case study of an example pipeline with a crack like defect shows that the failure pressure model plays a critical role in the risk management of pipelines through a life-cycle cost analysis.

Uncovering Bias: Exploring Machine Learning Techniques for Detecting and Mitigating Bias in Data – A Literature Review

The presence of Bias in models developed using machine learning algorithms has emerged as a critical issue. This literature review explores the topic of uncovering the existence of bias in data and the application of techniques for detecting and mitigating Bias. The review provides a comprehensive analysis of the existing literature, focusing on pre-processing techniques, post-pre-processing techniques, and fairness constraints employed to uncover and address the existence of Bias in machine learning models. The effectiveness, limitations, and trade-offs of these techniques are examined, highlighting their impact on advocating fairness and equity in decision-making processes.
 The methodology consists of two key steps: data preparation and bias analysis, followed by machine learning model development and evaluation. In the data preparation phase, the dataset is analyzed for biases and pre-processed using techniques like reweighting or relabeling to reduce bias. In the model development phase, suitable algorithms are selected, and fairness metrics are defined and optimized during the training process. The models are then evaluated using performance and fairness measures and the best-performing model is chosen. The methodology ensures a systematic exploration of machine learning techniques to detect and mitigate bias, leading to more equitable decision-making.
 The review begins by examining the techniques of pre-processing, which involve cleaning the data, selecting the features, feature engineering, and sampling. These techniques play an important role in preparing the data to reduce bias and promote fairness in machine learning models. The analysis highlights various studies that have explored the effectiveness of these techniques in uncovering and mitigating bias in data, contributing to the development of more equitable and unbiased machine learning models. Next, the review delves into post-pre-processing techniques that focus on detecting and mitigating bias after the initial data preparation steps. These techniques include bias detection methods that assess the disparate impact or disparate treatment in model predictions, as well as bias mitigation techniques that modify model outputs to achieve fairness across different groups. The evaluation of these techniques, their performance metrics, and potential trade-offs between fairness and accuracy are discussed, providing insights into the challenges and advancements in bias mitigation. Lastly, the review examines fairness constraints, which involve the imposition of rules or guidelines on machine learning algorithms to ensure fairness in predictions or decision-making processes. The analysis explores different fairness constraints, such as demographic parity, equalized odds, and predictive parity, and their effectiveness in reducing bias and advocating fairness in machine learning models. Overall, this literature review provides a comprehensive understanding of the techniques employed to uncover and mitigate the existence of bias in machine learning models. By examining pre-processing techniques, post-pre-processing techniques, and fairness constraints, the review contributes to the development of more fair and unbiased machine learning models, fostering equity and ethical decision-making in various domains. By examining relevant studies, this review provides insights into the effectiveness and limitations of various pre-processing techniques for bias detection and mitigation via Pre-processing, Adversarial learning, Fairness Constraints, and Post-processing techniques.

Brownfield redevelopment evaluation based on structure-process-outcome theory and continuous ordered weighted averaging operator-topology method

As an important part of urban renewal, brownfield restoration and renovation are of great significance to the sustainable development of cities. The structure-process-outcome theory was introduced into this study to improve the rationality and scientific vigor of the redevelopment assessment process and to evaluate whether brownfield sites meet the conditions for redevelopment. Based on this theory, the relationship among structures, processes and outcomes can be well elucidated. Specifically, a good structure should contribute to an effective process, which will increase the possibility of a favorable outcome. The basic conditions, practice principles, and result orientation in the whole procedure of brownfield redevelopment were comprehensively analyzed. In addition, a more complete and reasonable three-level evaluation index system for brownfield redevelopment was established. In order to reduce the subjectivity in the evaluation process, an unbiased scientific brownfield redevelopment evaluation model was constructed using the continuous ordered weighted averaging operator-topology method. The evaluation decision system was applied to the renovation of a tract project in Chengdu, China. The results proved that the model could effectively and accurately evaluate the quality level of the brownfield redevelopment project, and the proposed recommendations can provide a basis for decision-making.

Adapted Deep Ensemble Learning-Based Voting Classifier for Osteosarcoma Cancer Classification.

The study utilizes osteosarcoma hematoxylin and the Eosin-stained image dataset, which is unevenly dispersed, and it raises concerns about the potential impact on the overall performance and reliability of any analyses or models derived from the dataset. In this study, a deep-learning-based convolution neural network (CNN) and adapted heterogeneous ensemble-learning-based voting classifier have been proposed to classify osteosarcoma. The proposed methods can also resolve the issue and develop unbiased learning models by introducing an evenly distributed training dataset. Data augmentation is employed to boost the generalization abilities. Six different pre-trained CNN models, namely MobileNetV1, Mo-bileNetV2, ResNetV250, InceptionV2, EfficientNetV2B0, and NasNetMobile, are applied and evaluated in frozen and fine-tuned-based phases. In addition, a novel CNN model and adapted heterogeneous ensemble-learning-based voting classifier developed from the proposed CNN model, fine-tuned NasNetMobile model, and fine-tuned Efficient-NetV2B0 model are also introduced to classify osteosarcoma. The proposed CNN model outperforms other pre-trained models. The Kappa score obtained from the proposed CNN model is 93.09%. Notably, the proposed voting classifier attains the highest Kappa score of 96.50% and outperforms all other models. The findings of this study have practical implications in telemedicine, mobile healthcare systems, and as a supportive tool for medical professionals.

Distinguishing features of long COVID identified through immune profiling.

Post-acute infection syndromes may develop after acute viral disease1. Infection with SARS-CoV-2 can result in the development of a post-acute infection syndrome known as long COVID. Individuals with long COVID frequently report unremitting fatigue, post-exertional malaise, and a variety of cognitive and autonomic dysfunctions2-4. However, the biological processes that are associated with the development and persistence of these symptoms are unclear. Here 275 individuals with or without long COVID were enrolled in a cross-sectional study that included multidimensional immune phenotyping and unbiased machine learning methods to identify biological features associated with long COVID. Marked differences were noted in circulating myeloid and lymphocyte populations relative to the matched controls, as well as evidence of exaggerated humoral responses directed against SARS-CoV-2 among participants with long COVID. Furthermore, higher antibody responses directed against non-SARS-CoV-2 viral pathogens were observed among individuals with long COVID, particularly Epstein-Barr virus. Levels of soluble immune mediators and hormones varied among groups, with cortisol levels being lower among participants with long COVID. Integration of immune phenotyping data into unbiased machine learning models identified the key features that are most strongly associated with long COVID status. Collectively, these findings may help to guide future studies into the pathobiology of long COVID and help with developing relevant biomarkers.

Genomic prediction of seasonal forage yield in perennial ryegrass

AbstractBackgroundGenomic selection has the potential to accelerate genetic gain in perennial ryegrass breeding, provided complex traits such as forage yield can be predicted with sufficient accuracy.MethodsIn this study, we compared modelling approaches and feature selection strategies to evaluate the accuracy of genomic prediction models for seasonal forage yield production.ResultsOverall, model selection had limited impact on predictive ability when using the full data set. For a baseline genomic best linear unbiased prediction model, the highest mean predictive accuracy was obtained for spring grazing (0.78), summer grazing (0.62) and second cut silage (0.56). In terms of feature selection strategies, using uncorrelated single‐nucleotide polymorphisms (SNPs) had no impact on predictive ability, allowing for a potential decrease of the data set dimensions. With a genome‐wide association study, we found a significant SNP marker for spring grazing, located in the genic region annotated as coding for an enzyme responsible for fucosylation of xyloglucans—major components of the plant cell wall. We also presented an approach to increase interpretability of genomic prediction models with the use of Gene Ontology enrichment analysis.ConclusionsApproaches for feature selection will be relevant in development of low‐cost genotyping platforms in support of routine and cost‐effective implementation of genomic selection.

Basic-Class and Cross-Class Hybrid Feature Learning for Class-Imbalanced Weld Defect Recognition

Class-imbalanced weld defect recognition, which realizes defect recognition via learning features of class-imbalanced X-ray images, is an emerging but challenging task. Nevertheless, the existing studies on the class-imbalanced problem mainly focus on large-scale data, and it is difficult to extract high-quality features from the insufficient industrial data, resulting in weak recognition performance. To address the above issue, this paper comprehensively learns features from the perspective of basic-class and cross-class, on this basis, a novel hybrid feature learning model for class-imbalanced weld defect recognition is proposed. First, an image acquisition method completed by photographing, scanning and sampling is designed to collect the class-imbalanced X-ray images. Second, a hybrid feature learning model is proposed to learn the distinctive and effective features from acquired images, so that the class-imbalanced data is mapped to a balanced feature distribution. Third, with the distinguishable features learned by our hybrid feature learning model, an unbiased defect recognition model can be trained to recognize different types of defects. The practical weld data W-PPLN, W-MTL and W-GDXray are adopted in the experiments, and the experimental results show that our method outperforms the state-of-the-art (SOTA) methods on the task of class-imbalanced weld defect recognition.

A Recipe for Unbiased Background Modeling in Deep Wide-field Astronomical Images

Unbiased sky background modeling is crucial for the analysis of deep wide-field images, but it remains a major challenge in low surface brightness astronomy. Traditional image processing algorithms are often designed to produce artificially flat backgrounds, erasing astrophysically meaningful structures. In this paper, we present three ideas that can be combined to produce wide-field astronomical data that preserve accurate representations of the background sky: (1) Use of all-sky infrared/submillimeter data to remove the large-scale time-varying components while leaving the scattered light from Galactic cirrus intact, with the assumptions of (a) the underlying background has little power on small scales, and (b) the Galactic cirrus in the field is optically thin on large scales; (2) Censoring of frames contaminated by anomalously prominent wings in the wide-angle point-spread function; and (3) Incorporation of spatial covariance in image stacking that controls the local background consistency. We demonstrate these methods using example data sets obtained with the Dragonfly Telephoto Array, but these general techniques are prospective to be applied to improve sky models in data obtained from other wide-field imaging surveys, including those from the upcoming Vera Rubin Telescope.

A statistical package for evaluation of hybrid performance in plant breeding via genomic selection

Hybrid breeding employs heterosis, which could potentially improve the yield and quality of a crop. Genomic selection (GS) is a promising approach for the selection of quantitative traits in plant breeding. The main objectives of this study are to (i) propose a GS-based approach to identify potential parental lines and superior hybrid combinations from a breeding population, which is composed of hybrids produced by a half diallel mating design; (ii) develop a software package for users to carry out the proposed approach. An R package, designated EHPGS, was generated to facilitate the employment of the genomic best linear unbiased model considering additive plus dominance marker effects for the hybrid performance evaluation. The R package contains a Bayesian statistical algorithm for calculating genomic estimated breeding value (GEBVs), GEBV-based specific combining ability, general combining ability, mid-parent heterosis, and better-parent heterosis. Three datasets that have been published in literature, including pumpkin (Cucurbita maxima), maize (Zea mays), and wheat (Triticum aestivum L.), were reanalyzed to illustrate the use of EHPGS.

Ensemble of best linear unbiased predictor, machine learning and deep learning models predict maize yield better than each model alone

Abstract Predicting phenotypes accurately from genomic, environment and management factors is key to accelerating the development of novel cultivars with desirable traits. Inclusion of management and environmental factors enables in silico studies to predict the effect of specific management interventions or future climates. Despite the value such models would confer, much work remains to improve the accuracy of phenotypic predictions. Rather than advocate for a single specific modelling strategy, here we demonstrate within large multi-environment and multi-genotype maize trials that combining predictions from disparate models using simple ensemble approaches most often results in better accuracy than using any one of the models on their own. We investigated various ensemble combinations of different model types, model numbers and model weighting schemes to determine the accuracy of each. We find that ensembling generally improves performance even when combining only two models. The number and type of models included alter accuracy with improvements diminishing as the number of models included increases. Using a genetic algorithm to optimize ensemble composition reveals that, when weighted by the inverse of each model’s expected error, a combination of best linear unbiased predictor, linear fixed effects, deep learning, random forest and support vector regression models performed best on this dataset.

Application of single-step single nucleotide polymorphism best linear unbiased predictor model with unknown-parent groups for type traits in Japanese Holsteins

The objectives of this study were to investigate the computational performance and the predictive ability and bias of a single-step SNP BLUP model (ssSNPBLUP) in genotyped young animals with unknown-parent groups (UPG) for type traits, using national genetic evaluation data from the Japanese Holstein population. The phenotype, genotype, and pedigree data were the same as those used in a national genetic evaluation of linear type traits classified between April 1984 and December 2020. In the current study, 2 data sets were prepared: the full data set containing all entries up to December 2020 and a truncated data set ending with December 2016. Genotyped animals were classified into 3 types: sires with classified daughters (S), cows with records (C), and young animals (Y). The computing performance and prediction accuracy of ssSNPBLUP were compared for the following 3 groups of genotyped animals: sires with classified daughters and young animals (SY); cows with records and young animals (CY); and sires with classified daughters, cows with records, and young animals (SCY). In addition, we tested 3 parameters of residual polygenic variance in ssSNPBLUP (0.1, 0.2, or 0.3). Daughter yield deviations (DYD) for the validation bulls and phenotypes adjusted for all fixed effects and random effects other than animal and residual (Yadj) for the validation cows were obtained using the full data set from the pedigree-based BLUP model. The regression coefficients of DYD for bulls (or Yadj for cows) on the genomic estimated breeding value (GEBV) using the truncated data set were used to measure the inflation of the predictions of young animals. The coefficient of determination of DYD on GEBV was used to measure the predictive ability of the predictions for the validation bulls. The reliability of the predictions for the validation cows was calculated as the square of the correlation between Yadj and GEBV divided by heritability. The predictive ability was highest in the SCY group and lowest in the CY group. However, minimal difference was found in predictive abilities with or without UPG models using different parameters of residual polygenic variance. The regression coefficients approached 1.0 as the parameter of residual polygenic variance increased, but regression coefficients were mostly similar regardless of the use of UPG across the groups of genotyped animals. The ssSNPBLUP model, including UPG, was demonstrated as feasible for implementation in the national evaluation of type traits in Japanese Holsteins.

Local Phase Transitions in a Model of Multiplex Networks with Heterogeneous Degrees and Inter-Layer Coupling.

Multilayer networks represent multiple types of connections between the same set of nodes. Clearly, a multilayer description of a system adds value only if the multiplex does not merely consist of independent layers. In real-world multiplexes, it is expected that the observed inter-layer overlap may result partly from spurious correlations arising from the heterogeneity of nodes, and partly from true inter-layer dependencies. It is therefore important to consider rigorous ways to disentangle these two effects. In this paper, we introduce an unbiased maximum entropy model of multiplexes with controllable intra-layer node degrees and controllable inter-layer overlap. The model can be mapped to a generalized Ising model, where the combination of node heterogeneity and inter-layer coupling leads to the possibility of local phase transitions. In particular, we find that node heterogeneity favors the splitting of critical points characterizing different pairs of nodes, leading to link-specific phase transitions that may, in turn, increase the overlap. By quantifying how the overlap can be increased by increasing either the intra-layer node heterogeneity (spurious correlation) or the strength of the inter-layer coupling (true correlation), the model allows us to disentangle the two effects. As an application, we show that the empirical overlap observed in the International Trade Multiplex genuinely requires a nonzero inter-layer coupling in its modeling, as it is not merely a spurious result of the correlation between node degrees across different layers.

Uniform misclassification loss for unbiased model prediction

Deep learning algorithms have achieved tremendous success over the past few years. However, the biased behavior of deep models, where the models favor/disfavor certain demographic subgroups, is a major concern in the deep learning community. Several adverse consequences of biased predictions have been observed in the past. One solution to alleviate the problem is to train deep models for fair outcomes. Therefore, in this research, we propose a novel loss function, termed as Uniform Misclassification Loss (UML) to train deep models for unbiased outcomes. The proposed UML function penalizes the model for the worst-performing subgroup for mitigating bias and enhancing the overall model performance. The proposed loss function is also effective while training with imbalanced data as well. Further, a metric, Joint Performance Disparity Measure (JPD) is introduced to jointly measure the overall model performance and the bias in model prediction. Multiple experiments have been performed on four publicly available datasets for facial attribute prediction and comparisons are performed with existing bias mitigation algorithms. Experimental results are reported using performance and bias evaluation metrics. The proposed loss function outperforms existing bias mitigation algorithms that showcase its effectiveness in obtaining unbiased outcomes and improved performance.

Cooling streams with riparian trees: Thermal regime depends on total solar radiation penetrating the canopy

AbstractRiparian planting is often recommended for stream restoration, notably to mitigate solar heating of stream waters. However, plant leaves shade photosynthetically active radiation (PAR, 400–700 nm) more efficiently than the near‐infrared (NIR, 700–3000 nm) wavelengths that comprise about half of incident solar radiation and so contribute equally to water heating. Total solar radiation (NIR + PAR) exposure at the reach scale is needed to avoid bias when predicting the thermal response of streams. In this article, we alert stream ecologists and riparian restorers to the water heating contribution of NIR and provide a means to account for both NIR and PAR in total solar radiation penetrating riparian canopies. We used spectral pyranometers to simultaneously measure total solar radiation and its NIR component under different tree canopies over a wide range of shade conditions as indexed by PAR exposure. Measurements were made during full overcast (diffuse lighting) conditions so as to ‘sample’ transmission of radiation through the complete canopy. The NIR proportion of sub‐canopy solar radiation increased steadily with increasing shade and was appreciably greater under willow than pine canopies, while NZ native broadleaf rainforest had intermediate (rather variable) NIR content. Our trend lines for different tree canopies permit total sub‐canopy solar radiation to be estimated from PAR exposure for unbiased modelling of stream thermal regimes.

The potential of region-specific machine-learning-based ground motion models: Application to Turkey

Conventional ground motion models have extensively been established worldwide based on classical regression analysis of records. Alternatively, advanced nonparametric machine-learning (ML) algorithms may capture the complex nonlinear behaviour of earthquake motions. This paper investigates the efficiency of artificial neural network (ANN) and extreme gradient boosting (XGBoost) in predicting peak ground acceleration (PGA), peak ground velocity (PGV) and pseudo-spectral acceleration (PSA) (period, T = 0.03–2.0 s) for the Turkish dataset. The dataset involves 1166 records of 383 events with a moment magnitude (Mw) of 4.0–7.6, Joyner and Boore distance (RJB) of 0–200 km, focal depth (FD) less than 35 km, and site condition as the averaged shear wave velocity of the soil on the top 30 m (VS30) of 131–1380 m/s. The performance of the models is compared against empirical models in terms of root-mean-square error (RMSE), coefficient of determination (R2), Pearson correlation coefficient (r), and inter-event and intra-event residuals. To perform residual analysis, a likelihood function is developed. Findings reveal that the XGBoost approach gives an unbiased model with a higher correlation and lower residual than ANN. Finally, an online platform is provided for any interested users.

QSAR Modelling of Biological Activity in Cannabinoids with Quantum Similarity Combinations of Charge Fitting Schemes and 3D-QSAR.

Quantitative structure-activity relationship(QSAR) modeled the biological activities of 30 cannabinoids with quantum similarity descriptors(QSD) and Comparative Molecular Field Analysis (CoMFA). The PubChem[https://pubchem.ncbi.nlm.nih.gov/] database provided the geometries, binding affinities(Ki ) to the cannabinoid receptors type 1(CB1) and 2(CB2), and the median lethal dose(LD50 ) to breast cancer cells. An innovative quantum similarity approach combining (self)-similarity indexes calculated with different charge-fitting schemes under the Topo-Geometrical Superposition Algorithm(TGSA) were used to obtain QSARs. The determination coefficient(R2 ) and leave-one-out cross-validation[Q2 (LOO)] quantified the quality of multiple linear regression and support vector machine models. This approach was efficient in predicting the activities, giving predictive and robust models for each endpoint [pLD50 : R2 =0.9666 and Q2 (LOO)=0.9312; pKi (CB1): R2 =1.0000 and Q2 (LOO)=0.9727, and pKi (CB2): R2 =0.9996 and Q2 (LOO)=0.9460], where p is the negative logarithm. The descriptors based on the electrostatic potential encrypted better electronic information involved in the interaction. Moreover, the similarity-based descriptors generated unbiased models independent of an alignment procedure. The obtained models showed better performance than those reported in the literature. An additional 3D-QSAR CoMFA analysis was applied to 15 cannabinoids, taking THC as a template in a ligand-based approach. From this analysis, the region surrounding the amino group of the SR141716 ligand is the more favorable for the antitumor activity.

Detox Loss: Fairness Constraints for Learning With Imbalanced Data

Recent studies have highlighted a major caveat in several popular deep learning based face recognition algorithms. Despite their high accuracy, they have shown to exhibit biased behavior by achieving sub-par performance on images belonging to different sub-groups. Such behavior can often lead to unfair predictions, thus presenting a need to develop fairer models with unprejudiced behavior. To this effect, this research proposes Detox loss, a novel bias invariant feature learning loss function for learning unbiased models. The proposed loss can be used to: (i) learn fairer deep learning classifiers, and (ii) mitigate bias from existing pre-trained networks, especially in the challenging constraint of imbalanced training data with respect to a protected attribute. Conceptually, the Detox loss enforces that the learned features are distinguished based on the task label only, while eliminating any distinction based on the biasing-attribute. This is achieved by incorporating three fairness constraints while training with the traditional classification loss: (i) proposed Exclusion loss, (ii) proposed Inclusion loss, and (iii) proposed Feature-distillation loss. The efficacy of the Detox loss is demonstrated on two facial analysis tasks: (i) age-group prediction and (ii) gender prediction, under the protected attribute of ethnicity for learning fairer models and de-biasing existing models, with varying imbalanced training data distributions. Across two different protocols, three setups, and two tasks, the Detox loss obtains state of the art performance without the need of multi-labeled training data. For example, on the challenging Pilot Parliaments Benchmark (PPB) dataset, the Detox loss obtains a balanced accuracy and F1 score of 96.1% and 96.6%, as compared to the state-of-the-art performance of 94.1% and 93.6%, respectively.