Selection Gradients Research Articles

Jacobian norm with Selective Input Gradient Regularization for interpretable adversarial defense

Deep neural networks (DNNs) can be easily deceived by imperceptible alterations known as adversarial examples. These examples can lead to misclassification, posing a significant threat to the reliability of deep learning systems in real-world applications. Adversarial training (AT) is a popular technique used to enhance robustness by training models on a combination of corrupted and clean data. However, existing AT-based methods often struggle to handle transferred adversarial examples that can fool multiple defense models, thereby falling short of meeting the generalization requirements for real-world scenarios. Furthermore, AT typically fails to provide interpretable predictions, which are crucial for domain experts seeking to understand the behavior of DNNs. To overcome these challenges, we present a novel approach called Jacobian norm and Selective Input Gradient Regularization (J-SIGR). Our method leverages Jacobian normalization to improve robustness and introduces regularization of perturbation-based saliency maps, enabling interpretable predictions. By adopting J-SIGR, we achieve enhanced defense capabilities and promote high interpretability of DNNs. We evaluate the effectiveness of J-SIGR across various architectures by subjecting it to powerful adversarial attacks. Our experimental evaluations provide compelling evidence of the efficacy of J-SIGR against transferred adversarial attacks, while preserving interpretability. The project code can be found at https://github.com/Lywu-github/jJ-SIGR.git.

Machine Learning Big Data Analysis of the Impact of Air Pollutants on Rhinitis-Related Hospital Visits.

This study seeks to elucidate the intricate relationship between various air pollutants and the incidence of rhinitis in Seoul, South Korea, wherein it leveraged a vast repository of data and machine learning techniques. The dataset comprised more than 93 million hospital visits (n = 93,530,064) by rhinitis patients between 2013 and 2017. Daily atmospheric measurements were captured for six major pollutants: PM10, PM2.5, O3, NO2, CO, and SO2. We employed traditional correlation analyses alongside machine learning models, including the least absolute shrinkage and selection operator (LASSO), random forest (RF), and gradient boosting machine (GBM), to dissect the effects of these pollutants and the potential time lag in their symptom manifestation. Our analyses revealed that CO showed the strongest positive correlation with hospital visits across all three categories, with a notable significance in the 4-day lag analysis. NO2 also exhibited a substantial positive association, particularly with outpatient visits and hospital admissions and especially in the 4-day lag analysis. Interestingly, O3 demonstrated mixed results. Both PM10 and PM2.5 showed significant correlations with the different types of hospital visits, thus underlining their potential to exacerbate rhinitis symptoms. This study thus underscores the deleterious impacts of air pollution on respiratory health, thereby highlighting the importance of reducing pollutant levels and developing strategies to minimize rhinitis-related hospital visits. Further research considering other environmental factors and individual patient characteristics will enhance our understanding of these intricate dynamics.

Multi-locus genomic signatures of local adaptation to snow across the landscape in California populations of a willow leaf beetle.

Organisms living in mountains contend with extreme climatic conditions, including short growing seasons and long winters with extensive snow cover. Anthropogenic climate change is driving unprecedented, rapid warming of montane regions across the globe, resulting in reduced winter snowpack. Loss of snow as a thermal buffer may have serious consequences for animals overwintering in soil, yet little is known about how variability in snowpack acts as a selective agent in montane ecosystems. Here, we examine genomic variation in California populations of the leaf beetle Chrysomela aeneicollis, an emerging natural model system for understanding how organisms respond to climate change. We used a genotype-environment association approach to identify genomic signatures of local adaptation to microclimate in populations from three montane regions with variable snowpack and a coastal region with no snow. We found that both winter-associated environmental variation and geographical distance contribute to overall genomic variation across the landscape. We identified non-synonymous variation in novel candidate loci associated with cytoskeletal function, ion transport and membrane stability, cellular processes associated with cold tolerance in other insects. These findings provide intriguing evidence that variation in snowpack imposes selective gradients in montane ecosystems.

Digest: Multivariate genetic variation constrains adaptation to environmental changes during range expansion.

How do selection and standing genetic variation shape population divergence across landscapes? Henry and Stinchcombe estimated selection gradients on traits in the ivy-leaved morning glory (Ipomoea hederacea) in the field and compared them with the G-matrix and population divergence for four populations in North America. The authors show that population divergence and genetic covariances are largely unaligned with the selection gradient at the species' range edge. These findings raise the question of whether limited evolvability or multivariate genetic variation of populations at range edges prevent species from range expansion, which is important for understanding the role of genetic constraint in population divergence and predicting local adaptation in the face of climate change.

Sky Islands Are a Global Tool for Predicting the Ecological and Evolutionary Consequences of Climate Change

Sky islands are unique geologic formations, home to populations of organisms that have weathered climate change since the Pleistocene. Long-term isolation and climatic differences between sky islands and adjacent mountain chains result in natural laboratories well suited for examining the direct effects of climate change. Here, we review the global sky island literature to examine how taxa have responded to climate change. Results show lineage formation, reduced genetic variation, and trait evolution across taxa driven by genetic drift and natural selection. These effects continue today due to ongoing habitat reduction and steep selective gradients on sky islands relative to mountain chains. Comparative studies and experimental manipulations are needed to build broad inference into how past climate change has shaped the structure and function of whole ecosystems. The next era of sky island research is poised to create a model for climate change responses and eco-evolutionary dynamics, with profound conservation implications.

Population density affects sexual selection in an insect model

Abstract Understanding the impact of demographic and ecological factors on the strength of sexual selection is pivotal to predict eco‐evolutionary dynamics in populations facing environmental change. Density is expected to show pervasive intra‐ and interspecific variation in natural populations, but its effect on sexual selection remains largely unexplored. We tested experimentally the effect of density on standardized metrics of sexual selection in male and female red flour beetles. Specifically, we explored density effects during mating by manipulating the size of mating groups and the mating arena. We found that higher densities imposed by a larger group size amplified the opportunity for pre‐copulatory sexual selection in females and, to a smaller extent, also in males. Moreover, in males, we detected a higher opportunity for post‐copulatory sexual selection in larger groups. By contrast, higher densities imposed by differences in the size of the mating arena resulted in a steeper male sexual selection gradient. Overall, these data suggest that density is an important environmental factor contributing to the tremendous variation in the strength of sexual selection within and among species. Read the free Plain Language Summary for this article on the Journal blog.

Distinct spatial maps and multiple object codes in the lateral entorhinal cortex

The lateral entorhinal cortex (LEC) is a major cortical input area to the hippocampus, and it is crucial for associative object-place-context memories. An unresolved question is whether these associations are performed exclusively in the hippocampus or also upstream of it. Anatomical evidence suggests that the LEC processes both object and spatial information. We describe here a gradient of spatial selectivity along the antero-posterior axis of the LEC. We demonstrate that the LEC generates distinct spatial maps for different contexts that are independent of object coding and vice versa, thus providing evidence for pure spatial and pure object codes upstream of the hippocampus. While space and object coding occur by and large separately in the LEC, we identified neurons that encode for space and objects conjunctively. Together, these findings point to a scenario in which the LEC sustains both distinct space and object coding and associative space-object coding.

Phenotypic evolution of SARS-CoV-2: a statistical inference approach.

Since its emergence in late 2019, the SARS-CoV-2 virus has spread globally, causing the ongoing COVID-19 pandemic. In the fall of 2020, the Alpha variant (lineage B.1.1.7) was detected in England and spread rapidly, outcompeting the previous lineage. Yet, very little is known about the underlying modifications of the infection process that can explain this selective advantage. Here, we try to quantify how the Alpha variant differed from its predecessor on two phenotypic traits: the transmission rate and the duration of infectiousness. To this end, we analysed the joint epidemiological and evolutionary dynamics as a function of the Stringency Index, a measure of the amount of Non-Pharmaceutical Interventions. Assuming that these control measures reduce contact rates and transmission, we developed a two-step approach based on SEIR models and the analysis of a combination of epidemiological and evolutionary information. First, we quantify the link between the Stringency Index and the reduction in viral transmission. Secondly, based on a novel theoretical derivation of the selection gradient in an SEIR model, we infer the phenotype of the Alpha variant from its frequency changes. We show that its selective advantage is more likely to result from a higher transmission than from a longer infectious period. Our work illustrates how the analysis of the joint epidemiological and evolutionary dynamics of infectious diseases can help understand the phenotypic evolution driving pathogen adaptation.

Sex-specific heritabilities for length at maturity among Pacific salmonids and their consequences for evolution in response to artificial selection.

Artificial selection, whether intentional or coincidental, is a common result of conservation policies and natural resource management. To reduce unintended consequences of artificial selection, conservation practitioners must understand both artificial selection gradients on traits of interest and how those traits are correlated with others that may affect population growth and resilience. We investigate how artificial selection on male body size in Pacific salmon (Oncorhynchus spp.) may influence the evolution of female body size and female fitness. While salmon hatchery managers often assume that selection for large males will also produce large females, this may not be the case-in fact, because the fastest-growing males mature earliest and at the smallest size, and because female age at maturity varies little, small males may produce larger females if the genetic architecture of growth rate is the same in both sexes. We explored this possibility by estimating sex-specific heritability values of and natural and artificial selection gradients on length at maturity in four populations representing three species of Pacific salmon. We then used the multivariate breeder's equation to project how artificial selection against small males may affect the evolution of female length and fecundity. Our results indicate that the heritability of length at maturity is greater within than between the sexes and that sire-daughter heritability values are especially small. Salmon hatchery policies should consider these sex-specific quantitative genetic parameters to avoid potential unintended consequences of artificial selection.

Securing Distributed SGD Against Gradient Leakage Threats

This paper presents a holistic approach to gradient leakage resilient distributed Stochastic Gradient Descent (SGD). <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">First</i> , we analyze two types of strategies for privacy-enhanced federated learning: (i) gradient pruning with random selection or low-rank filtering and (ii) gradient perturbation with additive random noise or differential privacy noise. We analyze the inherent limitations of these approaches and their underlying impact on privacy guarantee, model accuracy, and attack resilience. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Next</i> , we present a gradient leakage resilient approach to securing distributed SGD in federated learning, with differential privacy controlled noise as the tool. Unlike conventional methods with the per-client federated noise injection and fixed noise parameter strategy, our approach keeps track of the trend of per-example gradient updates. It makes adaptive noise injection closely aligned throughout the federated model training. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Finally</i> , we provide an empirical privacy analysis on the privacy guarantee, model utility, and attack resilience of the proposed approach. Extensive evaluation using five benchmark datasets demonstrates that our gradient leakage resilient approach can outperform the state-of-the-art methods with competitive accuracy performance, strong differential privacy guarantee, and high resilience against gradient leakage attacks. The code associated with this paper can be found: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/git-disl/Fed-alphaCDP</uri> .

Predicting firm creation in rural Texas: A multi-model machine learning approach to a complex policy problem.

Rural and urban America have becoming increasingly divided, both politically and economically. Entrepreneurship can help rural communities catch back up by jumpstarting economic growth, creating jobs, and building resilience to economic shocks. However, less is known about firm creation in rural areas compared to urban areas. To that end, in this paper we ask: What factors predict firm creation in rural America? Our analysis, based on a comparative framework involving multiple machine learning modeling techniques, helps addresses three gaps in academic literature on rural firm creation. First, entrepreneurship research stretches across disciplines, often using econometric methods to identify the effect of a specific variable, rather than comparing the predictive importance of multiple variables. Second, research on firm creation centers on high-tech, urban firms. Third, modern machine learning techniques have not yet been applied in an integrated way to address rural entrepreneurship, a complex economic and policy problem that defies simple, monocausal claims. In this paper, we apply four machine learning methods (subset selection, lasso, random forest, and extreme gradient boosting) to a novel dataset to examine what social and economic factors are predictive of firm growth in rural Texas counties from 2008-2018. Our results suggest that some factors commonly discussed as promoting entrepreneurship (e.g., access to broadband and patents) may not be as predictive as socioeconomic ones (age distribution, ethnic diversity, social capital, and immigration). We also find that the strength of specific industries (oil, wind, healthcare, and elder/childcare) predicts firm growth, as does the number of local banks. Most factors predictive of firm growth in rural counties are distinct from those in urban counties, supporting the argument that rural entrepreneurship is a distinct phenomenon worthy of distinct focus. More broadly, this multi-model approach can offer initial, focusing guidance to policymakers seeking to address similarly complex policy problems.

Testing trade‐offs in amphibian metamorphic traits: Swimming performance and selection across metamorphic stages in the aquatic frog Xenopus laevis

Abstract As frogs undergo metamorphosis from tadpole to adult, they may be susceptible to decreases in locomotor performance and survival during stages in which the tail and limbs are both present. Because these alternative axial and appendicular propulsive systems might interfere with each other, performance might be better when only one or the other system predominates. This framework, termed the ‘adaptive peak hypothesis’ emerged from studies of generalist anuran taxa with biphasic life histories in which aquatic larvae transition to terrestrial adults. However, it is not clear that such patterns apply to taxa with different patterns of ontogenetic environmental transitions. We evaluated the relationship between morphology, locomotor performance and survival versus predators across metamorphic stages of a fully aquatic frog, the pipid Xenopus laevis. We measured escape performance from individuals of four developmental stages spanning possession of just a tail, the presence of both tail and legs and resorption of the tail, and then conducted selection trials in which equal numbers of individuals of all stages were exposed to an aquatic predator (cichlid fish). After collecting survivors from these trials, we calculated linear and nonlinear selection gradients for a suite of morphological traits. Unlike generalist frogs, we found that both locomotor performance and survival increased across metamorphic stages of X. laevis. Linear selection gradients favoured tadpoles with longer hindlimbs, longer and narrower bodies, and shorter tails—patterns that generally parallel trends during metamorphic growth. Nonlinear selection showed that longer hindlimbs were favoured as body length increased and the positive effect of forelimb length was reduced as tail length increased. The relationship between hindlimb and body length aligns with expected changes in growth postmetamorphosis where postmetamorphic individuals are expected to have greater survival. However, longer forelimbs are generally associated with decreased locomotor performance, suggesting that alternative mechanisms of selection may be acting on the forelimbs of Xenopus through development. Overall, our results indicate that the generality of the adaptive peak framework, as applied to amphibian metamorphosis, can be impacted by specifics of life history patterns across species. Read the free Plain Language Summary for this article on the Journal blog.

Exact expressions and numerical evaluation of average evolvability measures for characterizing and comparing textbf{G} matrices

Theory predicts that the additive genetic covariance (textbf{G}) matrix determines a population’s short-term (in)ability to respond to directional selection—evolvability in the Hansen–Houle sense—which is typically quantified and compared via certain scalar indices called evolvability measures. Often, interest is in obtaining the averages of these measures across all possible selection gradients, but explicit formulae for most of these average measures have not been known. Previous authors relied either on approximations by the delta method, whose accuracy is generally unknown, or Monte Carlo evaluations (including the random skewers analysis), which necessarily involve random fluctuations. This study presents new, exact expressions for the average conditional evolvability, average autonomy, average respondability, average flexibility, average response difference, and average response correlation, utilizing their mathematical structures as ratios of quadratic forms. The new expressions are infinite series involving top-order zonal and invariant polynomials of matrix arguments, and can be numerically evaluated as their partial sums with, for some measures, known error bounds. Whenever these partial sums numerically converge within reasonable computational time and memory, they will replace the previous approximate methods. In addition, new expressions are derived for the average measures under a general normal distribution for the selection gradient, extending the applicability of these measures into a substantially broader class of selection regimes.

Comparison of Machine Learning Algorithm Models in Bitcoin Price Sentiment Analysis

Bitcoin is one of the digital payments that is currently booming, fast delivery makes bitcoin in great demand by many people, currently there are many digital currency exchanges that can be used, one of the well-known ones in Indonesia, namely Indodax. Indodax is a cryptocurrency exchange, not only an exchange, Indodax also provides a chat room containing investors' opinions. Opinions contained in the Indodax chat room can be used to determine whether comments are positive, neutral or negative, so that it can be an investor's decision to sell or buy bitcoin using sentiment analysis. The sentiment analysis process begins with collecting data using an instant data scraper on the Indodax website, data preprcoessing, labeling using vader lexicon, TF-IDF as word weighting, data splitting, naïve Bayes algorithm and support vector machine, feature selection xgboost and gradient boosting, model evaluation with confusion matrix, then comparing the results of the two algorithms. Based on the tests that have been carried out, naïve bayes obtained the best accuracy value of 70.7%, naïve bayes combined with XGBoost obtained the best accuracy value of 86.6%, while the Support vector machine obtained the best accuracy 86.1%, support vector machine combined with gradient boosting obtained the best accuracy value of 88%. Based on these results the use of feature selection can increase the accuracy value of the algorithm.

Plastic responses to warmer climates: a semi-natural experiment on lizard populations.

Facing warming environments, species can exhibit plastic or microevolutionary changes in their thermal physiology to adapt to novel climates. Here, using semi-natural mesocosms, we experimentally investigated over two successive years whether a 2°C-warmer climate produces selective and inter- and intragenerational plastic changes in the thermal traits (preferred temperature and dorsal coloration) of the lizard Zootoca vivipara. In a warmer climate, the dorsal darkness, dorsal contrast, and preferred temperature of adults plastically decreased and covariances between these traits were disrupted. While selection gradients were overall weak, selection gradients for darkness were slightly different between climates and in the opposite direction to plastic changes. Contrary to adults, male juveniles were darker in warmer climates either through plasticity or selection and this effect was strengthened by intergenerational plasticity when juveniles' mothers also experienced warmer climates. While the plastic changes in adult thermal traits alleviate the immediate overheating costs of warming, its opposite direction to selective gradients and to juveniles' phenotypic responses may slow down evolutionary shifts toward phenotypes that are better adapted to future climates. Our study demonstrates the importance of considering inter- and intragenerational plasticity along with selective processes to better understand adaptation and population dynamics in light of climate change.

A novel lytic polysaccharide monooxygenase from enrichment microbiota and its application for shrimp shell powder biodegradation.

Lytic polysaccharide monooxygenases (LPMO) are expected to change the current status of chitin resource utilization. This study reports that targeted enrichment of the microbiota was performed with chitin by the selective gradient culture technique, and a novel LPMO (M2822) was identified from the enrichment microbiota metagenome. First, soil samples were screened based on soil bacterial species and chitinase biodiversity. Then gradient enrichment culture with different chitin concentrations was carried out. The efficiency of chitin powder degradation was increased by 10.67 times through enrichment, and chitin degradation species Chitiniphilus and Chitinolyticbacter were enriched significantly. A novel LPMO (M2822) was found in the metagenome of the enriched microbiota. Phylogenetic analysis showed that M2822 had a unique phylogenetic position in auxiliary activity (AA) 10 family. The analysis of enzymatic hydrolysate showed that M2822 had chitin activity. When M2822 synergized with commercial chitinase to degrade chitin, the yield of N-acetyl glycosamine was 83.6% higher than chitinase alone. The optimum temperature and pH for M2822 activity were 35°C and 6.0. The synergistic action of M2822 and chitin-degrading enzymes secreted by Chitiniphilus sp. LZ32 could efficiently hydrolyze shrimp shell powder. After 12 h of enzymatic hydrolysis, chitin oligosaccharides (COS) yield reached 4,724 μg/mL. To our knowledge, this work is the first study to mine chitin activity LPMO in the metagenome of enriched microbiota. The obtained M2822 showed application prospects in the efficient production of COS.

The magnitude of selection on growth varies among years and increases under warming conditions in a subarctic seabird.

Because of ongoing rapid climate change, many ecosystems are becoming both warmer and more variable, and these changes are likely to alter the magnitude and variability of natural selection acting on wild populations. Critically, changes and fluctuations in selection can impact both population demography and evolutionary change. Therefore, predicting the impacts of climate change depends on understanding the magnitude and variation in selection on traits across different life stages and environments. Long-term experiments in wild settings are a great opportunity to determine the impact of environmental conditions on selection. Here we examined variability in the strength of selection on size traits of nestling black-legged kittiwakes (Rissa tridactyla) in a 25-year study including a food supplementation experiment on Middleton Island in the Gulf of Alaska. Using mixed effect models, we examined the annual variability of stage-specific and resource-specific selection gradients across 25 years. We found that (a) larger and heavier hatchlings were the most likely to survive during early ontogeny, (b) non-food supplemented younger nestlings in a brood experienced the strongest selection, and (c) warmer conditions increased the magnitude of selection on nestling mass and affected non-food supplemented and second-hatched nestlings the most. Our results suggested that variable resource dynamics likely caused some of the changes in selection from year to year and that warming conditions increased the strength of selection on subarctic seabird growth. However, our experimental manipulation revealed that local environmental heterogeneity could buffer the selection expected from broader climatic changes. Consequently, understanding the interactive effects of local conditions and general changes in climate seems likely to improve our ability to predict future selection gradients.

Detecting frequency-dependent selection through the effects of genotype similarity on fitness components.

Frequency-dependent selection (FDS) is an evolutionary regime that can maintain or reduce polymorphisms. Despite the increasing availability of polymorphism data, few effective methods are available for estimating the gradient of FDS from the observed fitness components. We modeled the effects of genotype similarity on individual fitness to develop a selection gradient analysis of FDS. This modeling enabled us to estimate FDS by regressing fitness components on the genotype similarity among individuals. We detected known negative FDS on the visible polymorphism in a wild Arabidopsis and damselfly by applying this analysis to single-locus data. Further, we simulated genome-wide polymorphisms and fitness components to modify the single-locus analysis as a genome-wide association study (GWAS). The simulation showed that negative or positive FDS could be distinguished through the estimated effects of genotype similarity on simulated fitness. Moreover, we conducted the GWAS of the reproductive branch number in Arabidopsis thaliana and found that negative FDS was enriched among the top-associated polymorphisms of FDS. These results showed the potential applicability of the proposed method for FDS on both visible polymorphism and genome-wide polymorphisms. Overall, our study provides an effective method for selection gradient analysis to understand the maintenance or loss of polymorphism.

Exploration of novel biomarkers in Alzheimer's disease based on four diagnostic models.

Despite tremendous progress in diagnosis and prediction of Alzheimer's disease (AD), the absence of treatments implies the need for further research. In this study, we screened AD biomarkers by comparing expression profiles of AD and control tissue samples and used various models to identify potential biomarkers. We further explored immune cells associated with these biomarkers that are involved in the brain microenvironment. By differential expression analysis, we identified differentially expressed genes (DEGs) of four datasets (GSE125583, GSE118553, GSE5281, GSE122063), and common expression direction of genes of four datasets were considered as intersecting DEGs, which were used to perform enrichment analysis. We then screened the intersecting pathways between the pathways identified by enrichment analysis. DEGs in intersecting pathways that had an area under the curve (AUC) > 0.7 constructed random forest, least absolute shrinkage and selection operator (LASSO), logistic regression, and gradient boosting machine models. Subsequently, using receiver operating characteristic curve (ROC) and decision curve analysis (DCA) to select an optimal diagnostic model, we obtained the feature genes. Feature genes that were regulated by differentially expressed miRNAs (AUC > 0.85) were explored further. Furthermore, using single-sample GSEA to calculate infiltration of immune cells in AD patients. Screened 1855 intersecting DEGs that were involved in RAS and AMPK signaling. The LASSO model performed best among the four models. Thus, it was used as the optimal diagnostic model for ROC and DCA analyses. This obtained eight feature genes, including ATP2B3, BDNF, DVL2, ITGA10, SLC6A12, SMAD4, SST, and TPI1. SLC6A12 is regulated by miR-3176. Finally, the results of ssGSEA indicated dendritic cells and plasmacytoid dendritic cells were highly infiltrated in AD patients. The LASSO model is the optimal diagnostic model for identifying feature genes as potential AD biomarkers, which can supply new strategies for the treatment of patients with AD.

Quantitative Structure Retention-Relationship Modeling: Towards an Innovative General-Purpose Strategy.

Reversed-Phase Liquid Chromatography (RPLC) is a common liquid chromatographic mode used for the control of pharmaceutical compounds during their drug life cycle. Nevertheless, determining the optimal chromatographic conditions that enable this separation is time consuming and requires a lot of lab work. Quantitative Structure Retention Relationship models (QSRR) are helpful for doing this job with minimal time and cost expenditures by predicting retention times of known compounds without performing experiments. In the current work, several QSRR models were built and compared for their adequacy in predicting the retention times. The regression models were based on a combination of linear and non-linear algorithms such as Multiple Linear Regression, Support Vector Regression, Least Absolute Shrinkage and Selection Operator, Random Forest, and Gradient Boosted Regression. Models were built for five pH conditions, i.e., at pH 2.7, 3.5, 6.5, and 8.0. In the end, the model predictions were combined using stacking and the performances of all models were compared. The k-nearest neighbor-based application domain filter was established to assess the reliability of the prediction for further compound prioritization. Altogether, this study can be insightful for analytical chemists working with RPLC to begin with the computational prediction modeling such as QSRR to predict the separation of small molecules.