Latent Interaction Research Articles

A phenotypic drug discovery approach by latent interaction in deep learning

Contemporary drug discovery paradigms rely heavily on binding assays about the bio-physicochemical processes. However, this dominant approach suffers from overlooked higher-order interactions arising from the intricacies of molecular mechanisms, such as those involving cis -regulatory elements. It introduces potential impairments and restrains the potential development of computational methods. To address this limitation, I developed a deep learning model that leverages an end-to-end approach, relying exclusively on therapeutic information about drugs. By transforming textual representations of drug and virus genetic information into high-dimensional latent representations, this method evades the challenges arising from insufficient information about binding specificities. Its strengths lie in its ability to implicitly consider complexities such as epistasis and chemical–genetic interactions, and to handle the pervasive challenge of data scarcity. Through various modeling skills and data augmentation techniques, the proposed model demonstrates outstanding performance in out-of-sample validations, even in scenarios with unknown complex interactions. Furthermore, the study highlights the importance of chemical diversity for model training. While the method showcases the feasibility of deep learning in data-scarce scenarios, it reveals a promising alternative for drug discovery in situations where knowledge of underlying mechanisms is limited.

Tests of moderation of the big-fish-little-pond effect across students with and without special educational needs in regular versus special education schools

BackgroundInclusive education has become increasingly popular based on the assumption that it has multiple benefits for students with special educational needs (SEN). However, contradictions remain regarding the widespread nature of these benefits, particularly when it comes to academic motivation. AimsIn this large-scale cross-sectional study, we relied on the Big-Fish-Little-Pond Effect (BFLPE) to assess the links between inclusive education and students’ academic self-concepts and learning interests in the mathematics and verbal domains. SampleThe sample consisted of 21,219 German elementary school children in Grade 4 who were enrolled in three groups: students without SEN attending regular schools (n1 = 19,069), students with SEN attending regular schools (n2 = 933), and students with SEN attending special education schools (n3 = 1214). MethodsDoubly latent multi-group multilevel structural equation models and tests of latent interaction were performed to test the BFLPE. ResultsResults supported the BFLPE for both outcomes and domains. However, no BFLPE was identified for learning interest in the verbal domain among students with SEN attending special education schools, although the size of this effect did not differ significantly from that observed among students without SEN. In regular schools, the BFLPE was almost two times stronger among students with SEN than among their peers without SEN. ConclusionsResults support the generalizability of the BFLPE to students with SEN, while casting doubts on the motivational benefits of inclusive education for these students. Interventions targeted at attenuating the BFLPE should thus be tailored for both regular and special education schools.

LRAF-Net: Long-Range Attention Fusion Network for Visible-Infrared Object Detection.

Visible-infrared object detection aims to improve the detector performance by fusing the complementarity of visible and infrared images. However, most existing methods only use local intramodality information to enhance the feature representation while ignoring the efficient latent interaction of long-range dependence between different modalities, which leads to unsatisfactory detection performance under complex scenes. To solve these problems, we propose a feature-enhanced long-range attention fusion network (LRAF-Net), which improves detection performance by fusing the long-range dependence of the enhanced visible and infrared features. First, a two-stream CSPDarknet53 network is used to extract the deep features from visible and infrared images, in which a novel data augmentation (DA) method is designed to reduce the bias toward a single modality through asymmetric complementary masks. Then, we propose a cross-feature enhancement (CFE) module to improve the intramodality feature representation by exploiting the discrepancy between visible and infrared images. Next, we propose a long-range dependence fusion (LDF) module to fuse the enhanced features by associating the positional encoding of multimodality features. Finally, the fused features are fed into a detection head to obtain the final detection results. Experiments on several public datasets, i.e., VEDAI, FLIR, and LLVIP, show that the proposed method obtains state-of-the-art performance compared with other methods.

FoodAtlas: Automated knowledge extraction of food and chemicals from literature

Automated generation of knowledge graphs that accurately capture published information can help with knowledge organization and access, which have the potential to accelerate discovery and innovation. Here, we present an integrated pipeline to construct a large-scale knowledge graph using large language models in an active learning setting. We apply our pipeline to the association of raw food, ingredients, and chemicals, a domain that lacks such knowledge resources. By using an iterative active learning approach of 4120 manually curated premise-hypothesis pairs as training data for ten consecutive cycles, the entailment model extracted 230,848 food-chemical composition relationships from 155,260 scientific papers, with 106,082 (46.0 %) of them never been reported in any published database. To augment the knowledge incorporated in the knowledge graph, we further incorporated information from 5 external databases and ontology sources. We then applied a link prediction model to identify putative food-chemical relationships that were not part of the constructed knowledge graph. Validation of the 443 hypotheses generated by the link prediction model resulted in 355 new food-chemical relationships, while results show that the model score correlates well (R2 = 0.70) with the probability of a novel finding. This work demonstrates how automated learning from literature at scale can accelerate discovery and support practical applications through reproducible, evidence-based capture of latent interactions of diverse entities, such as food and chemicals.

TADGCN: A Time-Aware Dynamic Graph Convolution Network for long-term traffic flow prediction

Traffic flow prediction is a crucial component of the Intelligent Transportation System (ITS) with substantial implications for traffic management and daily travel. While short-term traffic forecasting has shown promise, long-term prediction presents unique challenges. The existing short-term approaches fall short in long-term scenarios due to two primary issues: (1) ignoring modeling the interaction relationships among dynamic spatial dependencies, and (2) extracting long-term temporal dependencies without capturing the time information in-depth. To mitigate the above issues, we propose a novel model called TADGCN, a Time-Aware Dynamic Graph Convolution Network, to effectively perform long-term traffic flow prediction. TADGCN captures latent interactions among dynamic spatial dependencies using Attention-Driven Dynamic Adaptive Graph Convolution Network (ADAGCN) modules. The time-aware capability is effectively enhanced using Time-Aware Joint Multi-View Temporal Encoder (TJMTE) modules due to the construction of a time-aware matrix. We conduct diversified experiments on three real-world datasets (England, PEMS04, and PEMS08) to evaluate the performance of our proposed method TADGCN. The results reveal that TADGCN achieves the best accuracy in all cases against 16 competitive state-of-the-art baselines.

Attitude, Habit Strength, and the Relationship Between Intention and Leisure-Time Physical Activity Behavior Among College Students

ABSTRACT According to the dual-process theories, many scholars proposed that physical activity behavior tends to be governed by an interplay of intentional planning and habitual action. The major purpose of this study was to determine the moderation effects of habit strength on the relationship between leisure time moderate-to-vigorous physical activity (MVPA) intention and behavior and the impacts of cognitive and affective attitude on PA habit strength, utilizing a latent interaction modeling approach. A total of 1803 undergraduate students completed this study. All variables were measured using validated self-report questionnaires. Structural equation modeling and Hayes’ PROCESS program (model 1) were conducted to address the research questions. Habit strength significantly and positively moderated the effects of intention on leisure time MVPA behavior (β = .08, p < .01). The association between intention and PA behavior became stronger as habit strength increased. Affective attitude positively and significantly influenced habit strength (β = .71, p < .01). The effect of cognitive attitude on habit strength was significant and negative with a small effect size (β = –.08, p < .01). Findings indicated that habit strength positively moderated the relationship between PA intention and behavior, suggesting that stronger habit strength may amplify the association between PA intention and actual PA behavior. Affective attitude toward PA significantly influenced habit strength, whereas cognitive attitude exhibited minimal negative impact on habit strength. These findings contribute to our understanding of the interaction effects between deliberative and automatic processes in regulating PA behavior, as well as the nuanced relationship between attitude and habit development.

Estimating nonlinear effects of random slopes: A comparison of multilevel structural equation modeling with a two-step, a single-indicator, and a plausible values approach

Multilevel structural equation modeling (MSEM) is a statistical framework of major relevance for research concerned with people’s intrapersonal dynamics. An application domain that is rapidly gaining relevance is the study of individual differences in the within-person association (WPA) of variables that fluctuate over time. For instance, an individual’s social reactivity – their emotional response to social situations – can be represented as the association between repeated measurements of the individual’s social interaction quantity and momentary well-being. MSEM allows researchers to investigate the associations between WPAs and person-level outcome variables (e.g., life satisfaction) by specifying the WPAs as random slopes in the structural equation on level 1 and using the latent representations of the slopes to predict outcomes on level 2. Here, we are concerned with the case in which a researcher is interested in nonlinear effects of WPAs on person-level outcomes – a U-shaped effect of a WPA, a moderation effect of two WPAs, or an effect of congruence between two WPAs – such that the corresponding MSEM includes latent interactions between random slopes. We evaluate the nonlinear MSEM approach for the three classes of nonlinear effects (U-shaped, moderation, congruence) and compare it with three simpler approaches: a simple two-step approach, a single-indicator approach, and a plausible values approach. We use a simulation study to compare the approaches on accuracy of parameter estimates and inference. We derive recommendations for practice and provide code templates and an illustrative example to help researchers implement the approaches.

Latent Interaction Effect in the CLPM Model: A Two-Step Multiple Imputation Analysis

This study aims to estimate the latent interaction effect in the CLPM model through a two-step multiple imputation analysis. The estimation of within × within and between × within latent interaction under the CLPM model framework is compared between the one-step Bayesian LMS method and the two-step multiple imputation analysis through a simulation study. The analysis show that the two-step multiple imputation analysis can provide unbiased estimation parameter, similar to the one-step Bayesian LMS method. This study also uses self-esteem and depression data from NLSY79 to perform a two-step multiple imputation analysis of the CLPM model, as well as an empirical example of latent interaction analysis. Mplus syntax is provided for researchers’ reference.

Graph attention network with convolutional layer for predicting gene regulations from single-cell ribonucleic acid sequence data

Reconstructing gene regulatory networks (GRNs) is an important task to reveal the regulatory relationship between genes and understand the mechanism of intracellular gene expression regulation. With the development of single-cell ribonucleic acid sequencing (scRNA-seq) technology, researchers begin to attempt to infer GRN within cells. In this paper, we propose graph attention network with convolutional layer (GATCL) to infer the latent interactions between transcription factors (TFs) and target genes in GRN. Firstly, GATCL uses graph attention network (GAT), which can effectively extract information about genes and TFs. Secondly, we combine multi-head attention layer with one-head attention layer and propose a new method of using convolution instead of weight matrix. Thirdly, we use the exponential linear unit (ELU) activation function to replace the leaky rectified linear unit (LReLU) commonly used in GAT, which further improves the accuracy of GATCL. The AUROC of our method on seven scRNA-seq datasets with four types of ground-truth networks reached an average of 0.827, which is higher than other state-of-the-art models. GATCL applies a supervised deep learning algorithm to solve the problems existing in the inference of GRN from scRNA-seq in the engineering field, and the effectiveness of this method is verified by a large number of experiments.

The synergy between stress and self‐compassion in building resilience: A 4‐year longitudinal study

AbstractThis 4‐year prospective study investigated the dynamic relationship between stress, self‐compassion, and resilience among university students, a population with increasing rates of mental health challenges. Drawing on stress theories, the research explored whether the combination of stress and self‐compassion strengthens resilience over time. A sample of 1137 university students (47.6% White, 38% female) completed measures of stress, resilience, and self‐compassion annually during the Spring semester across their four undergraduate years. A random intercept cross‐lagged panel model with latent interactions was used to test the hypothesized causal relationships. Contrary to the common belief that stress is debilitating, the results revealed a positive association between an increase in stress and a subsequent increase in self‐compassion. Moreover, when stress levels increased alongside self‐compassion, students demonstrated higher resilience. Notably, an increase in either stress or self‐compassion alone did not enhance resilience. These findings highlight the synergetic effect between self‐compassion and stress in enhancing resilience; under the right conditions, stress can lead to positive outcomes and increased capacity for coping with future stressors.

Investigating Latent Interaction Effects in Multiple-Group Analysis in the Structural Equation Modeling Framework

This study aims to address a gap in the social and behavioral sciences literature concerning interaction effects between latent factors in multiple-group analysis. By comparing two approaches for estimating latent interactions within multiple-group analysis frameworks using simulation studies and empirical data, we assess their relative merits. Our simulation study results demonstrated the superiority of the Latent Moderated Structural Equations (LMS) approach over the Unconstrained Product Indicator (UPI) method in power and providing more accurate estimates of differences in latent interactions between groups. Utilizing multiple-group analysis on real-world data, we evaluated its efficacy in identifying latent interaction differences between groups. Empirical data analysis did not reveal discrepancies between LMS and UPI in terms of detecting differences in latent interactions between boys and girls, although they indicated varying sizes of differences in interaction effects. The study concludes with recommendations for examining differences in latent interactions and suggests avenues for future research. Our findings aim to deepen the understanding of intricate relationships among variables, group differences, and the moderating effects of latent variables, ultimately facilitating the development of more accurate theoretical models.

Transformer fusion-based scale-aware attention network for multispectral victim detection

The aftermath of a natural disaster leaves victims trapped in rubble which is challenging to detect by smart drones due to the victims in low visibility under the adverse disaster environments and victims in various sizes. To overcome the above challenges, a transformer fusion-based scale-aware attention network (TFSANet) is proposed to overcome adverse environmental impacts in disaster areas by robustly integrating the latent interactions between RGB and thermal images and to address the problem of various-sized victim detection. Firstly, a transformer fusion model is developed to incorporate a two-stream backbone network to effectively fuse the complementary characteristics between RGB and thermal images. This aims to solve the problem that the victims cannot be seen clearly due to the adverse disaster area, such as smog and heavy rain. In addition, a scale-aware attention mechanism is designed to be embedded into the head network to adaptively adjust the size of receptive fields aiming to capture victims with different scales. Extensive experiments on two challenging datasets indicate that our TFSANet achieves superior results. The proposed method achieves 86.56% average precision (AP) on the National Institute of Informatics—Chiba University (NII-CU) multispectral aerial person detection dataset, outperforming the state-of-the-art approach by 4.38%. On the drone-captured RGBT person detection (RGBTDronePerson) dataset, the proposed method significantly improves the AP of the state-of-the-art approach by 4.33%.

Multi-modal trajectory forecasting with Multi-scale Interactions and Multi-pseudo-target Supervision

Trajectory forecasting is crucial for the advancement of autonomous vehicles. While much progress has been made, extant approaches often fall short in accounting for intricate social interactions and the unpredictability of human behavior. This paper introduces a novel multi-modal trajectory forecasting model named Multi-scale Interaction and Multi-pseudo-target Supervision (MIMS). Central to our approach is a multiscale hypergraph that discerns latent interactions across all scales, evaluating both their strength and type. Further enhancing our model’s capabilities is a causal-effect-estimation-based pseudo-target generation method. This facilitates multi-modality modeling in motion forecasting by offering explicit supervision with multiple latent targets. We validate MIMS using the Argoverse Motion Forecasting dataset. The results reveal that MIMS outperforms current state-of-the-art models across various traffic conditions. Specifically, our multiscale hypergraph exhibits a superior capacity for capturing complex spatiotemporal dependencies compared to pairwise methods. Moreover, our multi-pseudo-target supervision eliminates the pattern convergence of multi-modal forecasting.

Identifying latent genetic interactions in genome-wide association studies using multiple traits

The "missing" heritability of complex traits may be partly explained by genetic variants interacting with other genes or environments that are difficult to specify, observe, and detect. We propose a new kernel-based method called Latent Interaction Testing (LIT) to screen for genetic interactions that leverages pleiotropy from multiple related traits without requiring the interacting variable to be specified or observed. Using simulated data, we demonstrate that LIT increases power to detect latent genetic interactions compared to univariate methods. We then apply LIT to obesity-related traits in the UK Biobank and detect variants with interactive effects near known obesity-related genes (URL: https://CRAN.R-project.org/package=lit).

Daily stress reactivity and risk appraisal mediates childhood parental abuse predicting adulthood psychopathology severity: An 18-year longitudinal mediation analysis

Identifying mechanisms of childhood abuse-adulthood psychopathology relations could facilitate preventive efforts, but most prior studies used cross-sectional or two-wave designs and did not test the effects of childhood maternal and paternal abuse separately. Our 18-year three-wave study thus determined if Wave 2 daily stress reactivity and risk appraisal severity mediated Wave 1 retrospectively-reported childhood maternal and paternal abuse on Wave 3 generalized anxiety disorder (GAD), major depressive disorder (MDD), panic disorder (PD), alcohol (AUD), and substance use disorder (SUD) self-rated symptom severity. Longitudinal structural equation modeling was employed, adjusting for Wave 1 psychopathology severity. Higher childhood maternal and paternal abuse consistently predicted greater future daily stress reactivity and risk appraisal, and these mediators subsequently predicted increased GAD, MDD, and PD, but not AUD and SUD severity. Daily stress reactivity and risk appraisal consistently mediated the pathways between childhood maternal and paternal abuse predicting heightened adulthood GAD, MDD, and PD (Cohen's d = 0.333–0.888) but not AUD and SUD severity. Mediation effect sizes were stronger for childhood maternal (24.5–83.0%) than paternal (19.5–56.0%) abuse as the predictor. The latent interaction between Wave 1 childhood maternal and paternal abuse did not moderate the effect of Wave 1 maternal or paternal abuse on any Wave 3 adulthood psychopathology severity through Wave 2 daily stress reactivity and risk appraisal. Our research emphasizes the urgent requirement for continuous evaluation and intervention initiatives in trauma-informed care, both in inpatient and outpatient treatment settings.

Anti-lung cancer synergy of low-dose doxorubicin and PD-L1 blocker co-delivered via mild photothermia-responsive black phosphorus.

We have previously identified a latent interaction mechanism between non-small cell lung cancer cells (NSCLCC) and their associated macrophages (TAM) mediated by mutual paracrine activation of the HMGB1/RAGE/NF-κB signaling. Activation of this mechanism results in TAM stimulation and PD-L1 upregulation in the NSCLCC. In the present work, we found that free DOX at a low concentration that does not cause DNA damage could activate the HMGB1/RAGE/NF-κB/PD-L1 pathway byinducing oxidative stress. It was thus proposed that a combination of low-dose DOX and a PD-L1 blocker delivered in the NSCLC tumor would achieve synergistic TAM stimulation and thereby synergetic anti-tumor potency. To prove this idea, DOX and BMS-202 (a PD-L1 blocker) were loaded to black phosphorus (BP) nanoparticles after dosage titration to yield the BMS-202/DOX@BP composites that rapidly disintegrated and released drug cargo upon mild photothermal heating at 40°C. In vitro experiments then demonstrated that low-dose DOX and BMS-202 delivered via BMS-202/DOX@BP under mild photothermia displayed enhanced tumor cell toxicity with a potent synergism only in the presence of TAM. This enhanced synergism was due to an anti-tumor M1-like TAM phenotype that was synergistically induced by low dose DOX plus BMS-202 only in the presence of the tumor cells, indicating the damaged tumor cells to be the cardinal contributor to the M1-like TAM stimulation. In vivo, BMS-202/DOX@BP under mild photothermia exhibited targeted delivery to NSCLC graft tumors in mice and synergistic anti-tumor efficacy of delivered DOX and BMS-202. In conclusion, low-dose DOX in combination with a PD-L1 blocker is an effective strategy to turn TAM against their host tumor cells exploiting the HMGB1/RAGE/NF-κB/PD-L1 pathway. The synergetic actions involved highlight the value of TAM and the significance of modulating tumor cell-TAM cross-talk in tumor therapy. Photothermia-responsive BP provides an efficient platform to translate this strategy into targeted, efficacious tumor therapy.

On Representation Knowledge Distillation for Graph Neural Networks.

Knowledge distillation (KD) is a learning paradigm for boosting resource-efficient graph neural networks (GNNs) using more expressive yet cumbersome teacher models. Past work on distillation for GNNs proposed the local structure preserving (LSP) loss, which matches local structural relationships defined over edges across the student and teacher's node embeddings. This article studies whether preserving the global topology of how the teacher embeds graph data can be a more effective distillation objective for GNNs, as real-world graphs often contain latent interactions and noisy edges. We propose graph contrastive representation distillation (G-CRD), which uses contrastive learning to implicitly preserve global topology by aligning the student node embeddings to those of the teacher in a shared representation space. Additionally, we introduce an expanded set of benchmarks on large-scale real-world datasets where the performance gap between teacher and student GNNs is non-negligible. Experiments across four datasets and 14 heterogeneous GNN architectures show that G-CRD consistently boosts the performance and robustness of lightweight GNNs, outperforming LSP (and a global structure preserving (GSP) variant of LSP) as well as baselines from 2-D computer vision. An analysis of the representational similarity among teacher and student embedding spaces reveals that G-CRD balances preserving local and global relationships, while structure preserving approaches are best at preserving one or the other.

Supplemental Material for Normality Assumption in Latent Interaction Models

Supplemental Material for Normality Assumption in Latent Interaction Models

Cognitive Reappraisal Moderates the Longitudinal Relationship between Adolescents' Peer Victimization and Self-Esteem. A Latent Interaction Model.

Poor self-esteem relates closely to youth maladjustment and appears to be predicted by peer victimization experiences. However, not all peer victimized adolescents face the same risk for self-esteem erosion over time. Drawing upon the Bi-Dimensional Framework for resilience and extant research, the present study examined the potential moderating role of cognitive reappraisal in the prospective relationship from peer victimization to self-esteem. To increase precision of findings the long-term impact of self-esteem on peer victimization was also tested. Self-reported data were collected from 285 early adolescents (Mage = 10.53 years, SD = 0.16; 54.0% girls) at two waves, spaced 1-year. Latent moderated structural equation analysis showed that peer victimization was negatively related to later self-esteem, but only for youth displaying low levels of cognitive reappraisal. For adolescents with high levels of cognitive reappraising, peer victimization was not found to predict any changes in self-esteem over time. The long-term impact of self-esteem on peer victimization was not supported. Overall the present study suggests that enhancing cognitive reappraisal could be a promising avenue for lowering risk for poor self-esteem in young individuals experiencing peer victimization.

The moderating role of personality in the job strain process: A latent interaction approach

The moderating role of personality in the job strain process: A latent interaction approach