Spatial Datasets Research Articles

A robust approach to Gaussian process implementation

Abstract. Gaussian process (GP) regression is a flexible modeling technique used to predict outputs and to capture uncertainty in the predictions. However, the GP regression process becomes computationally intensive when the training spatial dataset has a large number of observations. To address this challenge, we introduce a scalable GP algorithm, termed MuyGPs, which incorporates nearest-neighbor and leave-one-out cross-validation during training. This approach enables the evaluation of large spatial datasets with state-of-the-art accuracy and speed in certain spatial problems. Despite these advantages, conventional quadratic loss functions used in the MuyGPs optimization, such as root mean squared error (RMSE), are highly influenced by outliers. We explore the behavior of MuyGPs in cases involving outlying observations and, subsequently, develop a robust approach to handle and mitigate their impact. Specifically, we introduce a novel leave-one-out loss function based on the pseudo-Huber function (LOOPH) that effectively accounts for outliers in large spatial datasets within the MuyGPs framework. Our simulation study shows that the LOOPH loss method maintains accuracy despite outlying observations, establishing MuyGPs as a powerful tool for mitigating unusual observation impacts in the large data regime. In the analysis of US ozone data, MuyGPs provides accurate predictions and uncertainty quantification, demonstrating its utility in managing data anomalies. Through these efforts, we advance the understanding of GP regression in spatial contexts.

SDFSD-v1.0: A Sub-Meter SAR Dataset for Fine-Grained Ship Detection

In the field of target detection, a prominent area is represented by ship detection in SAR imagery based on deep learning, particularly for fine-grained ship detection, with dataset quality as a crucial factor influencing detection accuracy. Datasets constructed with commonly used slice-based annotation methods suffer from a lack of scalability and low efficiency in repeated editing and reuse. Existing SAR ship datasets mostly consist of medium to low resolution imagery, leading to coarse ship categories and limited background scenarios. We developed the “annotate entire image, then slice” workflow (AEISW) and constructed a sub-meter SAR fine-grained ship detection dataset (SDFSD) by using 846 sub-meter SAR images that include 96,921 ship instances of 15 ship types across 35,787 slices. The data cover major ports and shipping routes globally, with varied and complex backgrounds, offering diverse annotation information. Several State-of-the-Art rotational detection models were used to evaluate the dataset, providing a baseline for ship detection and fine-grained ship detection. The SDFSD is a high spatial resolution ship detection dataset that could drive advancements in research on ship detection and fine-grained detection in SAR imagery.

Enhancing spatial domain detection in spatial transcriptomics with EnSDD

Advancements in spatial transcriptomics have transformed our understanding of organ function and tissue microenvironment. However, accurately identifying spatial domains to depict genome heterogeneity and cellular interactions remains a challenge. In this study, we propose EnSDD (Ensemble-learning for Spatial Domain Detection), a method that ingeniously integrates eight state-of-the-art spatial domain detection methods to automatically identify spatial domains. A key innovation of EnSDD is its dynamic weighting mechanism within the ensemble learning process, which optimizes the contribution of each base model and provides a performance evaluation metric without the need for ground truth data. By leveraging the spatial domains identified through EnSDD, we incorporate the detection of domain-specific spatially variable genes and the spatial distribution of cell types, thereby providing deeper insights into tissue heterogeneity. We validate EnSDD across diverse spatial transcriptomics datasets from various tissue organizational structures. Our results demonstrate that EnSDD significantly enhances spatial domain identification accuracy, identifies genes with spatial expression patterns, and reveals domain-specific cell type enrichment patterns, offering invaluable insights into tissue spatial heterogeneity and regionalization.

A General Method for Resampling Autocorrelated Spatial Data

AbstractComparing spatial data sets is a ubiquitous task in data analysis, however the presence of spatial autocorrelation means that standard estimates of variance will be wrong and tend to over‐estimate the statistical significance of correlations and other observations. While there are a number of existing approaches to this problem, none are ideal, requiring detailed analytical calculations, which are hard to generalize or detailed modeling of the data generating process, which may not be straightforward. In this work we propose an approach based on permuting or resampling at fixed spatial autocorrelation, measured by Moran's I, in order to generate a null model that accounts for spatial dependence. Testing on real and synthetic data, we find that, as long as the spatial autocorrelation is not too strong, this approach works just as well as if we knew the data generating process exactly and allows us to compute P‐values with the correct Type‐I error rate.

ScTML: a pan-cancer single-cell landscape of multiple mutation types.

Investigating mutations, including single nucleotide variations (SNVs), gene fusions, alternative splicing and copy number variations (CNVs), is fundamental to cancer study. Recent computational methods and biological research have demonstrated the reliability and biological significance of detecting mutations from single-cell transcriptomic data. However, there is a lack of a single-cell-level database containing comprehensive mutation information in all types of cancer. Establishing a single-cell mutation landscape from the huge emerging single-cell transcriptomic data can provide a critical resource for elucidating the mechanisms of tumorigenesis and evolution. Here, we developed scTML (http://sctml.xglab.tech/), the first database offering a pan-cancer single-cell landscape of multiple mutation types. It includes SNVs, insertions/deletions, gene fusions, alternative splicing and CNVs, along with gene expression, cell states and other phenotype information. The data are from 74 datasets with 2 582 633 cells, including 35 full-length (Smart-seq2) transcriptomic single-cell datasets (all publicly available data with raw sequencing files), 23 datasets from 10X technology and 16 spatial transcriptomic datasets. scTML enables users to interactively explore multiple mutation landscapes across tumors or cell types, analyze single-cell-level mutation-phenotype associations and detect cell subclusters of interest. scTML is an important resource that will significantly advance deciphering intra-tumor and inter-tumor heterogeneity, and how mutations shape cell phenotypes.

Enhancing local-scale groundwater quality predictions using advanced machine learning approaches

Assessing groundwater quality typically involves labor-intensive, time-consuming, and costly laboratory tests, making real-time monitoring impractical, especially at the local level. Groundwater quality projections at the local scale using broad spatial datasets have been inaccurate due to variations in hydrogeology, human activities, industrial operations, groundwater extraction, and waste disposal. This study aims to identify the most dependable and resilient machine learning algorithms for forecasting groundwater quality at nearby monitoring locations by utilizing simple water quality metrics that can be quickly assessed without extensive sampling and laboratory testing. The Entropy-weighted Water Quality Index (EWQI) was calculated using a large spatial and temporal dataset (2014–2021) of 977 wells with parameters including pH, total hardness (TH), calcium (Ca2⁺), magnesium (Mg2⁺), sodium (Na⁺), potassium (K⁺), sulfate (SO₄2⁻), chloride (Cl⁻), nitrate (NO₃⁻), total dissolved solids (TDS), and fluoride (F⁻). Further, similar parameters were also observed in 33 open wells at the three local monitoring sites from December 2022 to March 2023. The EWQI was predicted using a Random Forest (RF), eXtreme Gradient Boosting (XGB), and Deep Neural Network (DNN). The pH, TH, and TDS were used as input variables for EWQI predictions, as they can be easily measured using handheld probes or multi-parameters. The model performance was evaluated using R2, MAE, and RMSE. During the training stage, all three models predicted the EWQI with an R2 greater than 90%, with minimal errors when pH, TH, and TDS were input variables. To validate the models at a local scale, the EWQI was predicted at the village level (e.g., Antoli, Balapura, and Lapodiaya) using pH, TH, and TDS as input variables. The machine learning models were able to predict the EWQI very closely to the actual EWQI, with an R2 greater than 90%. It is also evident that the models could predict the EWQI using basic parameters that are easily measured, providing an overall idea of the water quality for a small area. Hence, these machine learning models could be useful for accurately representing groundwater quality, thereby avoiding the use of time-consuming and costly laboratory techniques.

Measuring Environmental and Behavioral Drivers of Chronic Diseases Using Smartphone-Based Digital Phenotyping: Intensive Longitudinal Observational mHealth Substudy Embedded in 2 Prospective Cohorts of Adults.

Previous studies investigating environmental and behavioral drivers of chronic disease have often had limited temporal and spatial data coverage. Smartphone-based digital phenotyping mitigates the limitations of these studies by using intensive data collection schemes that take advantage of the widespread use of smartphones while allowing for less burdensome data collection and longer follow-up periods. In addition, smartphone apps can be programmed to conduct daily or intraday surveys on health behaviors and psychological well-being. The aim of this study was to investigate the feasibility and scalability of embedding smartphone-based digital phenotyping in large epidemiological cohorts by examining participant adherence to a smartphone-based data collection protocol in 2 ongoing nationwide prospective cohort studies. Participants (N=2394) of the Beiwe Substudy of the Nurses' Health Study 3 and Growing Up Today Study were followed over 1 year. During this time, they completed questionnaires every 10 days delivered via the Beiwe smartphone app covering topics such as emotions, stress and enjoyment, physical activity, access to green spaces, pets, diet (vegetables, meats, beverages, nuts and dairy, and fruits), sleep, and sitting. These questionnaires aimed to measure participants' key health behaviors to combine them with objectively assessed high-resolution GPS and accelerometer data provided by participants during the same period. Between July 2021 and June 2023, we received 11.1 TB of GPS and accelerometer data from 2394 participants and 23,682 survey responses. The average follow-up time for each participant was 214 (SD 148) days. During this period, participants provided an average of 14.8 (SD 5.9) valid hours of GPS data and 13.2 (SD 4.8) valid hours of accelerometer data. Using a 10-hour cutoff, we found that 51.46% (1232/2394) and 53.23% (1274/2394) of participants had >50% of valid data collection days for GPS and accelerometer data, respectively. In addition, each participant submitted an average of 10 (SD 11) surveys during the same period, with a mean response rate of 36% across all surveys (SD 17%; median 41%). After initial processing of GPS and accelerometer data, we also found that participants spent an average of 14.6 (SD 7.5) hours per day at home and 1.6 (SD 1.6) hours per day on trips. We also recorded an average of 1046 (SD 1029) steps per day. In this study, smartphone-based digital phenotyping was used to collect intensive longitudinal data on lifestyle and behavioral factors in 2 well-established prospective cohorts. Our assessment of adherence to smartphone-based data collection protocols over 1 year suggests that adherence in our study was either higher or similar to most previous studies with shorter follow-up periods and smaller sample sizes. Our efforts resulted in a large dataset on health behaviors that can be linked to spatial datasets to examine environmental and behavioral drivers of chronic disease.

Non-stationary mechanical sound source separation: An all-neural beamforming network driven by time–frequency convolution and self-attention

To tackle the challenging task of extracting target signal features amidst significant interference from complex, time-varying non-stationary noise in industrial scenarios, we propose a novel time-domain all-neural beamforming network tailored for non-stationary mechanical sound source separation, named TFANBNet. Leveraging time–frequency convolution and self-attention, TFANBNet promised robust performance in challenging acoustic environments. First, the proposed time–frequency convolution module employs parameterized complex-valued convolutional layers to simulate time–frequency transformations, thereby implicitly extracting time–frequency features. Then, the proposed adaptive attention transformer network enhances local attention interaction by integrating intermediate features, effectively enhancing long-range context modeling. Third, the beamforming module utilizes deep neural networks for beamforming weight estimation, replacing conventional noise covariance matrix computations, which have inherent performance limitations. Finally, the experimental results on the multi-channel spatial reverberation dataset synthesized from the MIMII dataset show that the proposed TFANBNet yields superior separation performance and generalization capabilities compared to the considered competitive methods.

Rapid Appraisal of Wildlife Corridor Viability with Geospatial Modelling and Field Data: Lessons from Makuyuni, Tanzania

Connectivity between protected areas is necessary to prevent habitat fragmentation. Biodiverse countries like Tanzania craft legislation to promote habitat connectivity via the creation of ecological corridors, but their viability for wildlife often remains unknown. We therefore develop a scalable and replicable approach to assess and monitor multispecies corridor viability using geospatial modeling and field data. We apply and test the approach in the Makuyuni study area: an unprotected ecological corridor connecting Tarangire National Park to Essmingor mountain, Makuyuni Wildlife Park and Mto Wa Mbu Game Controlled Area. We analyzed the viability of Makuyuni as an ecological corridor by creating and validating a geospatial least-cost corridor model with field observations of wildlife and livestock. We created the model from publicly available spatial datasets augmented with manual digitization of pastoral homesteads (bomas). The least-cost corridor model identified two likely pathways for wildlife, confirmed and validated with field observations. Locations with low least-cost values were significantly correlated with more wildlife observations (Spearman’s rho = −0.448, p = 0.002). Our findings suggest that Makuyuni is a viable ecological corridor threatened by development and land use change. Our methodology presents a replicable approach for both monitoring Makuyuni and assessing corridor viability more generally. The incorporation of manually digitized homesteads (bomas) and field-based livestock observations makes corridor assessment more robust by taking into account pastoral land uses that are often missing in land cover maps. Integration of geospatial analysis and field observations is key for the robust identification of corridors for habitat connectivity.

Insights into Boreal Forest Disturbance from Canopy Stability Index

The world’s forests are being increasingly disturbed from exposure to the compounding impacts of land use and climate change, in addition to natural disturbance regimes. Boreal forests have a lower level of deforestation compared to tropical forests, and while they have higher levels of natural disturbances, the accumulated impact of forest management for commodity production coupled with worsening fire weather conditions and other climate-related stressors is resulting in ecosystem degradation and loss of biodiversity. We used satellite-based time-series analysis of two canopy indices—canopy photosynthesis and canopy water stress—to calculate an index that maps the relative stability of forest canopies in the Canadian provinces of Ontario and Quebec. By drawing upon available spatial time-series data on logging, wildfire, and insect infestation impacts, we were able to attribute the causal determinants of areas identified as having unstable forest canopy. The slope of the two indices that comprise the stability index also provided information as to where the forest is recovering from prior disturbances. The stability analyses and associated spatial datasets are available in an interactive web-based mapping app. that can be used to map disturbed forest canopies and the attribution of disturbances to human or natural causes. This information can assist decision-makers in identifying areas that are potentially ecologically degraded and in need of restoration and those stable areas that are a priority for protection.

Imputing spatial transcriptomics through gene network constructed from protein language model.

Image-based spatial transcriptomic sequencing technologies have enabled the measurement of gene expression at single-cell resolution, but with a limited number of genes. Current computational approaches attempt to overcome these limitations by imputing missing genes, but face challenges regarding prediction accuracy and identification of cell populations due to the neglect of gene-gene relationships. In this context, we present stImpute, a method to impute spatial transcriptomics according to reference scRNA-seq data based on the gene network constructed from the protein language model ESM-2. Specifically, stImpute employs an autoencoder to create gene expression embeddings for both spatial transcriptomics and scRNA-seq data, which are used to identify the nearest neighboring cells between scRNA-seq and spatial transcriptomics datasets. According to the neighbored cells, the gene expressions of spatial transcriptomics cells are imputed through a graph neural network, where nodes are genes, and edges are based on cosine similarity between the ESM-2 embeddings of the gene-encoding proteins. The gene prediction uncertainty is further measured through a deep learning model. stImpute was shown to consistently outperform state-of-the-art methods across multiple datasets concerning imputation and clustering. stImpute also demonstrates robustness in producing consistent results that are insensitive to model parameters.

HSPEI: A 1‐km spatial resolution SPEI dataset across the Chinese mainland from 2001 to 2022

HSPEI: A 1‐km spatial resolution SPEI dataset across the Chinese mainland from 2001 to 2022

A Comprehensive Spatially Resolved Metabolomics Dataset for Lampreys

As one of the two most ancient groups of extant vertebrates, lamprey has become an important model organism in various fields of biology. In this paper, we present a comprehensive tissue-wide spatial metabolomics dataset for lampreys, where 14 distinct tissues were analyzed using liquid chromatography-mass spectrometry (LC-MS) in both positive and negative ion modes. The dataset has been fully validated using internal standard and pooled quality control samples and is readily accessible at the UCSD Metabolomics Workbench. This dataset serves as a valuable resource for researchers using lampreys as a model organism. Additionally, it acts as a benchmark metabolomics dataset for evaluating new algorithms and software tools and comparing them with previously published results. A lamprey spatial metabolomics database is also provided to support studies utilizing lampreys as an animal model, and to complement and validate other spatial metabolomics studies on lampreys conducted with mass spectrometry imaging or other techniques.

Toward reproducible and interoperable environmental modeling: Integration of HydroShare with server-side methods for exposing large-extent spatial datasets to models

Reproducible environmental modelling often relies on spatial datasets as inputs, typically manually subset for specific areas. Yet, models can benefit from a data distribution approach facilitated by online repositories, and automating processes to foster reproducibility. This study introduces a method leveraging diverse state-scale spatial datasets to create cohesive packages for GIS-based environmental modelling. These datasets were generated and shared via GeoServer and THREDDS Data Server connected to HydroShare, contrasting with conventional distribution methods. Using the Regional Hydro-Ecologic Simulation System (RHESSys) across three U.S. catchment-scale watersheds, we demonstrate minimal errors in spatial inputs and model streamflow outputs compared to traditional approaches. This spatial data-sharing method facilitates consistent model creation, fostering reproducibility. Its broader impact allows scientists to tailor the method to various use cases, such as exploring different scales beyond state-scale or applying it to other online repositories using existing data distribution systems, eliminating the need to develop their own.

ChatGeoAI: Enabling Geospatial Analysis for Public through Natural Language, with Large Language Models

Large Language Models (LLMs) such as GPT, BART, and Gemini stand at the forefront of Generative Artificial Intelligence, showcasing remarkable prowess in natural language comprehension and task execution. This paper proposes a novel framework developed on the foundation of Llama 2, aiming to bridge the gap between natural language queries and executable code for geospatial analyses within the PyQGIS environment. It empowers non-expert users to leverage GIS technology without requiring deep knowledge of geospatial programming or tools. Through cutting-edge Natural Language Processing (NLP) techniques, including tailored entity recognition and ontology mapping, the framework accurately interprets user intents and translates them into specific GIS operations. Integration of geospatial ontologies enriches semantic comprehension, ensuring precise alignment between user descriptions, geospatial datasets, and geospatial analysis tasks. A code generation module empowered by Llama 2 converts these interpretations into PyQGIS scripts, enabling the execution of geospatial analysis and results visualization. Rigorous testing across a spectrum of geospatial analysis tasks, with incremental complexity, evaluates the framework and the performance of such a system, with LLM at its core. The proposed system demonstrates proficiency in handling various geometries, spatial relationships, and attribute queries, enabling accurate and efficient analysis of spatial datasets. Moreover, it offers robust error-handling mechanisms and supports tasks related to map styling, visualization, and data manipulation. However, it has some limitations, such as occasional struggles with ambiguous attribute names and aliases, which leads to potential inaccuracies in the filtering and retrieval of features. Despite these limitations, the system presents a promising solution for applications integrating LLMs into GIS and offers a flexible and user-friendly approach to geospatial analysis.

Evaluation of the space-time effects of Covid-19 on household loans and savings in Romania - A spatial panel data approach at county level

This article examines the effects of the Covid-19 crisis on the banking sector of Romania using a spatial panel dataset of 21 months, from April 2020 to December 2021, among 41 counties. The empirical setup exploits the spatial and time-series variations of the banking sector during the pandemic via means of exploratory data analysis, simple linear models for each month with spatial diagnosis, and spatial panel data models with fixed effects. County-level differences and particular disturbances in the trend of banking services are highlighted against the turning points of the pandemic. We show that increasing Covid-19 infections result in higher loans and savings per capita, with significant spatial interactions given by the neighboring counties.

INSPIRE: interpretable, flexible and spatially-aware integration of multiple spatial transcriptomics datasets from diverse sources.

Recent advances in spatial transcriptomics technologies have led to a growing number of diverse datasets, offering unprecedented opportunities to explore tissue organizations and functions within spatial contexts. However, it remains a significant challenge to effectively integrate and interpret these data, often originating from different samples, technologies, and developmental stages. In this paper, we present INSPIRE, a deep learning method for integrative analyses of multiple spatial transcriptomics datasets to address this challenge. With designs of graph neural networks and an adversarial learning mechanism, INSPIRE enables spatially informed and adaptable integration of data from varying sources. By incorporating non-negative matrix factorization, INSPIRE uncovers interpretable spatial factors with corresponding gene programs, revealing tissue architectures, cell type distributions and biological processes. We demonstrate the capabilities of INSPIRE by applying it to human cortex slices from different samples, mouse brain slices with complementary views, mouse hippocampus and embryo slices generated through different technologies, and spatiotemporal organogenesis atlases containing half a million spatial spots. INSPIRE shows superior performance in identifying detailed biological signals, effectively borrowing information across distinct profiling technologies, and elucidating dynamical changes during embryonic development. Furthermore, we utilize INSPIRE to build 3D models of tissues and whole organisms from multiple slices, demonstrating its power and versatility.

Single-cell multi-omics map of human fetal blood in Down syndrome

Down syndrome predisposes individuals to haematological abnormalities, such as increased number of erythrocytes and leukaemia in a process that is initiated before birth and is not entirely understood1–3. Here, to understand dysregulated haematopoiesis in Down syndrome, we integrated single-cell transcriptomics of over 1.1 million cells with chromatin accessibility and spatial transcriptomics datasets using human fetal liver and bone marrow samples from 3 fetuses with disomy and 15 fetuses with trisomy. We found that differences in gene expression in Down syndrome were dependent on both cell type and environment. Furthermore, we found multiple lines of evidence that haematopoietic stem cells (HSCs) in Down syndrome are ‘primed’ to differentiate. We subsequently established a Down syndrome-specific map linking non-coding elements to genes in disomic and trisomic HSCs using 10X multiome data. By integrating this map with genetic variants associated with blood cell counts, we discovered that trisomy restructured regulatory interactions to dysregulate enhancer activity and gene expression critical to erythroid lineage differentiation. Furthermore, as mutations in Down syndrome display a signature of oxidative stress4,5, we validated both increased mitochondrial mass and oxidative stress in Down syndrome, and observed that these mutations preferentially fell into regulatory regions of expressed genes in HSCs. Together, our single-cell, multi-omic resource provides a high-resolution molecular map of fetal haematopoiesis in Down syndrome and indicates significant regulatory restructuring giving rise to co-occurring haematological conditions.

Scale matters: How spatial resolution impacts remote sensing based urban green space mapping?

Urban green spaces (UGS) provide ecological and habitat benefits such as carbon sequestration, oxygen production, humidity increase, noise reduction, and pollution absorption. UGS maps derived from remote sensing images serve as the fundamental data for urban planning and carbon sequestration assessments. However, the spatial resolution of remote sensing image and the pattern of urban structures significantly influence UGS mapping, making it challenging to obtain accurate UGS maps. To investigate the impact of spatial resolution on UGS mapping, this study utilized five different spatial resolution datasets: Gaofen2 (1 m, 4 m), Sentinel2 (10 m), and Landsat8 (15 m, 30 m). Random forest, LightGBM, and support vector machine were employed to map UGS, and the accuracies of UGS maps at different spatial resolutions were compared. Subsequently, the spatial distribution patterns of uncertainties in UGS maps were analyzed from both overall and urban functional zone perspectives. Furthermore, the uncertainty analysis of UGS mapping was conducted considering different landscape patterns in urban functional zones. The results indicate: (1) UGS map varies at different spatial resolution. Higher uncertainties associated with coarser spatial resolutions. Medium and coarse spatial resolution images inadequately capture the fine-grained distribution of urban green spaces. (2) Uncertainty in UGS mapping at different spatial resolutions is generally consistent in spatial distribution. From a functional zoning perspective, the accuracy of green space mapping over non-natural zones is sensitive to spatial resolution. (3) The distribution pattern of UGS patches affects the accuracy of UGS mapping. Uncertainty can be reduced in UGS mapping at medium and coarse spatial resolutions based on UGS landscape pattern indices by multiple linear regression, random forest and LightGBM model. This study comprehensively reveals that uncertainties in mapping UGS from multi-spatial resolution remote sensing images vary across urban functional zones and landscape pattern indices, and it is the first attempt to propose methods for UGS area correction based on landscape pattern indices. The results of this study will facilitate the application of remote sensing data at different spatial resolutions in urban areas.

ELLA: Modeling Subcellular Spatial Variation of Gene Expression within Cells in High-Resolution Spatial Transcriptomics.

Spatial transcriptomic technologies are becoming increasingly high-resolution, enabling precise measurement of gene expression at the subcellular level. Here, we introduce a computational method called subcellular expression localization analysis (ELLA), for modeling the subcellular localization of mRNAs and detecting genes that display spatial variation within cells in high-resolution spatial transcriptomics. ELLA creates a unified cellular coordinate system to anchor diverse cell shapes and morphologies, utilizes a nonhomogeneous Poisson process to model spatial count data, leverages an expression gradient function to characterize subcellular expression patterns, and produces effective control of type I error and high statistical power. We illustrate the benefits of ELLA through comprehensive simulations and applications to four spatial transcriptomics datasets from distinct technologies, where ELLA not only identifies genes with distinct subcellular localization patterns but also associates these patterns with unique mRNA characteristics. Specifically, ELLA shows that genes enriched in the nucleus exhibit an abundance of long noncoding RNAs or protein-coding mRNAs, often characterized by longer gene lengths. Conversely, genes containing signal recognition peptides, encoding ribosomal proteins, or involved in membrane related activities tend to enrich in the cytoplasm or near the cellular membrane. Furthermore, ELLA reveals dynamic subcellular localization patterns during the cell cycle, with certain genes showing decreased nuclear enrichment in the G1 phase while others maintain their enrichment patterns throughout the cell cycle. Overall, ELLA represents a calibrated, powerful, robust, scalable, and versatile tool for modeling subcellular spatial expression variation across diverse high-resolution spatial transcriptomic platforms.