Dissecting Spatiotemporal Structures in Spatial Transcriptomics via Diffusion-Based Adversarial Learning.

Rapid advances in spatial transcriptomics (ST) have revolutionized the interrogation of spatial heterogeneity and increase the demand for comprehensive methods to effectively characterize spatial domains. As a prerequisite for ST data analysis, spatial domain characterization is a crucial step for downstream analyses and biological implications. Here we propose a prior-based self-attention framework for spatial transcriptomics (PAST), a variational graph convolutional autoencoder for ST, which effectively integrates prior information via a Bayesian neural network, captures spatial patterns via a self-attention mechanism, and enables scalable application via a ripple walk sampler strategy. Through comprehensive experiments on data sets generated by different technologies, we show that PAST can effectively characterize spatial domains and facilitate various downstream analyses, including ST visualization, spatial trajectory inference and pseudotime analysis. Also, we highlight the advantages of PAST for multislice joint embedding and automatic annotation of spatial domains in newly sequenced ST data. Compared with existing methods, PAST is the first ST method that integrates reference data to analyze ST data. We anticipate that PAST will open up new avenues for researchers to decipher ST data with customized reference data, which expands the applicability of ST technology.

Dimension reduction and (spatial) clustering is usually performed sequentially; however, the low-dimensional embeddings estimated in the dimension-reduction step may not be relevant to the class labels inferred in the clustering step. We therefore developed a computation method, Dimension-Reduction Spatial-Clustering (DR-SC), that can simultaneously perform dimension reduction and (spatial) clustering within a unified framework. Joint analysis by DR-SC produces accurate (spatial) clustering results and ensures the effective extraction of biologically informative low-dimensional features. DR-SC is applicable to spatial clustering in spatial transcriptomics that characterizes the spatial organization of the tissue by segregating it into multiple tissue structures. Here, DR-SC relies on a latent hidden Markov random field model to encourage the spatial smoothness of the detected spatial cluster boundaries. Underlying DR-SC is an efficient expectation-maximization algorithm based on an iterative conditional mode. As such, DR-SC is scalable to large sample sizes and can optimize the spatial smoothness parameter in a data-driven manner. With comprehensive simulations and real data applications, we show that DR-SC outperforms existing clustering and spatial clustering methods: it extracts more biologically relevant features than conventional dimension reduction methods, improves clustering performance, and offers improved trajectory inference and visualization for downstream trajectory inference analyses.

Spatial transcriptomics (ST) technologies enable dissecting the tissue architecture in spatial context. To perceive the global contextual information of gene expression patterns in tissue, the spatial dependence of cells must be fully considered by integrating both local and non-local features by means of spatial-context-aware. However, the current ST integration algorithm ignores for ST dropouts, which impedes the spatial-aware of ST features, resulting in challenges in the accuracy and robustness of microenvironmental heterogeneity detecting, spatial domain clustering, and batch-effects correction. Here, we developed an STMGraph, a universal dual-view dynamic deep learning framework that combines dual-remask (MASK-REMASK) with dynamic graph attention model (DGAT) to exploit ST data outperforming pre-existing tools. The dual-remask mechanism masks the embeddings before encoding and decoding, establishing dual-decoding-view to share features mutually. DGAT leverages self-supervision to update graph linkage relationships from two distinct perspectives, thereby generating a comprehensive representation for each node. Systematic benchmarking against 10 state-of-the-art tools revealed that the STMGraph has the optimal performance with high accuracy and robustness on spatial domain clustering for the datasets of diverse ST platforms from multi- to sub-cellular resolutions. Furthermore, STMGraph aggregates ST information cross regions by dual-remask to realize the batch-effects correction implicitly, allowing for spatial domain clustering of ST multi-slices. STMGraph is platform independent and superior in spatial-context-aware to achieve microenvironmental heterogeneity detection, spatial domain clustering, batch-effects correction, and new biological discovery, and is therefore a desirable novel tool for diverse ST studies

TOGAR: Token-gated generative refinement for high-fidelity spatial transcriptomics and robust spatial domain clustering.

Spatial transcriptomics is a rapidly developing field of single-cell genomics that quantitatively measures gene expression while providing spatial information within tissues. A key challenge in spatial transcriptomics is identifying spatially structured domains, which involves analyzing transcriptomic data to find clusters of cells with similar expression patterns and their spatial distribution. To address these challenges, we propose a novel deep-learning method called DMGCN for domain identification. The process begins with preprocessing that constructs two types of graphs: a spatial graph based on Euclidean distance and a feature graph based on Cosine distance. These graphs represent spatial positions and gene expressions, respectively. The embeddings of both graphs are generated using a multi-view graph convolutional encoder with an attention mechanism, enabling separate and co-convolution of the graphs, as well as corrupted feature convolution for contrastive learning. Finally, a fully connected network (FCN) decoder is employed to generate domain labels and reconstruct gene expressions for downstream analysis. Experimental results demonstrate that DMGCN consistently outperforms state-of-the-art methods in various tasks, including spatial clustering, trajectory inference, and gene expression broadcasting.

<p dir="ltr">Early diagnosis and effective treatment of complex and malignant diseases are critical challenges in modern medicine. These diseases exhibit significant heterogeneity, manifesting differently across patients and involving diverse genes and pathways. This variability complicates the identification of universal biomarkers, necessitating innovative, system-level approaches to uncover robust and predictive biomarkers across various biological contexts.</p><p dir="ltr">This dissertation addresses these challenges by leveraging both knowledge- based and data-driven methods to analyze multi-omics data, including spatial and single-cell transcriptomics, as well as bulk data. The goal is to prioritize potential biomarkers that can provide system-level insights into disease mechanisms, ultimately facilitating personalized medicine. The findings of this thesis promise to transform biomarker discovery, leading to earlier diagnoses, improved treatment outcomes, and enhanced patient care.</p><p dir="ltr">The thesis encompasses four comprehensive studies:</p><p dir="ltr">1. Multi-organ single-cell analysis reveals an on/off switch system with potential for personalized treatment of immunological diseases: We analyzed scRNA-seq data from mice and humans with immune-mediated inflammatory diseases (IMIDs). We identified upstream regulators acting as on/off switches, which collectively serve as candidate biomarkers and potential drug targets. These findings were validated in a study involving sera from nearly 300 systemic lupus erythematosus patients.</p><p dir="ltr">2. An interactive atlas of genomic, proteomic, and metabolomic biomarkers promotes the potential of proteins to predict complex diseases: Using multiomics data from UK Biobank (UKBB), we identified reliable biomarkers for nine complex diseases and constructed an interactive atlas to explore their performance. Proteomics outperformed other omics types in most of the diseases.</p><p dir="ltr">3. Multiomics biomarkers were not superior to clinical variables for pan- cancer screening: We analyzed multiomics data to find early diagnostic biomarkers for cancers. While proteomics data from peripheral blood did not outperform routine clinical variables for most cancers, promising biomarkers were identified for cancers in highly vascularized organs.</p><p dir="ltr">4. Combining spatial transcriptomics and pseudotime to find urine biomarkers for prostate cancer: We analyzed spatial transcriptomics data of prostate cancer to identify transcripts correlated with malignant transformation. Using pseudotime and machine learning, we validated the diagnostic accuracy of these transcripts in various samples from over 2,000 patients and controls.</p><p dir="ltr">In conclusion, this thesis demonstrates the potential of systems-level approaches in biomarker discovery, providing insights into disease mechanisms and paving the way for personalized medicine. The studies highlight the importance of considering disease heterogeneity and the need for tailored biomarker strategies, ultimately contributing to earlier diagnoses and improved patient outcomes.</p><h3>List of scientific papers</h3><p dir="ltr">I. Sandra Lilja*, Xinxiu Li*, <b>Martin Smelik</b>*, Eun Jung Lee, Joseph Loscalzo, Pratheek Bellur Marthanda, Lang Hu, Mattias Magnusson, Oleg Sysoev, Huan Zhang, Yelin Zhao, Christopher Sjöwall, Danuta Gawel, Hui Wang, Mikael Benson. Multi-organ single-cell analysis reveals an on/off switch system with potential for personalized treatment of immunological diseases. Cell Rep Med. 2023 Mar 21;4(3):100956. <a href="https://doi.org/10.1016/j.xcrm.2023.100956" rel="noreferrer" target="_blank">https://doi.org/10.1016/j.xcrm.2023.100956</a></p><p dir="ltr">II. <b>Martin Smelik</b>*, Yelin Zhao*, Xinxiu Li, Joseph Loscalzo, Oleg Sysoev, Firoj Mahmud, Dina Mansour Aly, Mikael Benson. An interactive atlas of genomic, proteomic, and metabolomic biomarkers promotes the potential of proteins to predict complex diseases. Sci Rep. 14, 12710 (2024). <a href="https://doi.org/10.1038/s41598-024-63399-9" rel="noreferrer" target="_blank">https://doi.org/10.1038/s41598-024-63399-9</a></p><p dir="ltr">III. <b>Martin Smelik</b>, Yelin Zhao, Dina Mansour Aly, AKM Firoj Mahmud, Oleg Sysoev, Xinxiu Li, Mikael Benson. Multiomics biomarkers were not superior to clinical variables for pan-cancer screening. Communications Medicine. 4, 234 (2024). <a href="https://doi.org/10.1038/s43856-024-00671-z" rel="noreferrer" target="_blank">https://doi.org/10.1038/s43856-024-00671-z</a></p><p dir="ltr">IV. <b>Martin Smelik</b>*, Daniel Diaz-Roncero Gonzalez*, Xiaojing An, Rakesh Heer, Lars Henningsohn, Xinxiu Li, Hui Wang, Yelin Zhao, Mikael Benson. Combining spatial transcriptomics, pseudotime and machine learning to find biomarkers for prostate cancer. [Manuscript]</p><p dir="ltr">* Shared first-author</p>

The application of machine learning (ML) to genomics has transformed the process of analyzing and interpreting large-scale, complex datasets, leading to important breakthroughs in our knowledge of biological systems. This review provides a comprehensive overview of ML applications in key genomic areas: Whole Genome Sequencing (WGS), Whole Exome Sequencing (WES), single-cell genomics, and spatial transcriptomics. In WGS and WES, ML techniques are employed for variant calling, genome-wide association studies, rare variant analysis, and the prediction of pathogenicity. In single-cell genomics, ML facilitates clustering, trajectory inference, and cell type identification, while in spatial transcriptomics, it aids in deciphering spatial patterns of gene expression and tissue heterogeneity. This review further explores the application of ML in related omics fields, including proteomics, transcriptomics, metagenomics, epigenomics, and microbiome research. These applications encompass protein structure prediction, functional annotation, microbial community profiling, and the analysis of epigenetic modifications. We address the challenges caused by high dimensionality, variability in the data, and the requirement for interpretable machine learning models when dealing with genomic data. Emerging technologies like explainable AI and federated learning are highlighted for their potential to address these challenges. Additionally, the review addresses ethical considerations, data privacy issues, and the necessity for standardized protocols in ML applications. This comprehensive examination underscores the transformative impact of ML in genomics and highlights its potential to drive future innovations in personalized medicine and biological research. Received: 17 August 2024 | Revised: 17 October 2024 | Accepted: 31 October 2024 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support this work are available upon reasonable request to the corresponding author. Author Contribution Statement Saheed Adegbola Adeyanju: Conceptualization, Methodology, Validation, Data curation, Writing - original draft, Writing - review & editing, Supervision. Taiwo Temitope Ogunjobi: Methodology, Investigation, Writing - original draft, Writing - review & editing.

Hepatocellular carcinoma (HCC) is one of the most common types of cancer globally. However, HCC features poor prognosis due to complex pathogenesis and limitations of therapeutic approaches. To improve the prognosis of HCC patients, we analyzed single-cell transcriptome and spatial transcriptome on liver tissues from HCC patients. We applied Non-negative Matrix Factorization (NMF) method on spatial transcriptome and found a critical pathway associated with HCC through GO over-representation analysis. According to the pathway activity score, the cells in single-cell transcriptome were divided into three groups. Then we conducted hdWGCNA and selected HCC-related co-expressed gene modules which showed significant intergroup differences. Integrative machine learning algorithms on bulk transcriptome were employed for core genes selection. We respectively evaluated their AUC values as independent diagnostic markers and retained a core gene. We conducted a survival analysis using this gene. Next, we performed immune infiltration analysis, tumor microenvironmental analysis and drug sensitivity analysis. Finally, pseudotime analysis, differentiation potential analysis and cell-cell communication analysis were applied to decipher the mechanisms underlying tumor initiation and progression. Single-cell transcriptome analysis revealed the high heterogeneity of cells in patients with HCC. The GO over-representation analysis based on NMF indicated that cytoplasmic translation path was enriched in most NMF components. The SSR2 gene passed the integrative machine learning algorithms and possessed excellent diagnosis (AUC > 0.8) and prognostic abilities (K-M curve, P = 0.007). Further analysis identified SSR2 as the top-ranked gene by average expression level in NK cells compared to all other immune cell clusters. Its expression showed a consistent downward trend during NK cell maturation.NK cells with high expression of SSR2 exhibited stronger cell-cell communication than those with low expression of SSR2. NK cells mainly communicated with other cells through MIF-mediated signaling pathway. In hepatocellular carcinoma (HCC), the high expression of the SSR2 gene in NK cells is accompanied by enhanced cytoplasmic translation pathway, which promotes MIF secretion and activates its downstream signaling pathways, thereby driving the disruption of the tumor immune microenvironment and the progression of the disease.

The rapid advancement of spatial transcriptomics (ST) sequencing technology has made it possible to capture gene expression with spatial coordinate information at the cellular level. Although many methods in ST data analysis can detect spatially variable genes (SVGs), these methods often fail to identify genes with explicit spatial expression patterns due to the lack of consideration for spatial domains. Considering spatial domains is crucial for identifying SVGs as it focuses the analysis of gene expression changes on biologically relevant regions, aiding in the more accurate identification of SVGs associated with specific cell types. Existing methods for identifying SVGs based on spatial domains predefine spot similarity before training, which prevents adaptive learning and limits generalizability across different tissues or samples. This limitation may also lead to inaccurate identification of specific genes at boundary regions. To address these issues, we present GAADE, an unsupervised neural network architecture based on graph-structured data representation learning. GAADE stacks encoder/decoder layers and integrates a self-attention mechanism to reconstruct node attributes and graph structure, effectively capturing spatial domain structures of different sections. Consequently, we confine the identification of SVGs within spatial domains. By performing differential expression analysis on spots within the target spatial domain and their multi-order neighbors, GAADE detects genes with enriched expression patterns within defined domains. Comparative evaluations with five other popular methods on ST datasets across four different species, regions and tissues demonstrate that GAADE exhibits superior performance in detecting SVGs and capturing the extent of spatial gene expression variation.

ABSTRACTDespite enabling high‐resolution mapping of gene expression within tissues, spatial transcriptomics (ST) still faces challenges in accurately segmenting spatial domains due to complex tissue architecture and limitations of current methods. Most approaches rely on local spatial priors, lack gene‐level interpretability, and fall short in capturing structure‐discriminative genes or long‐range functional relationships, limiting their ability to resolve biologically meaningful architectures. We present Spatially Aware Gene selection and dual‐view Embedding fusion (SAGE), a unified and reproducible framework for domain identification in spatial transcriptomics that combines topic‐driven gene selection with dual‐view embedding fusion to address these gaps. SAGE integrates non‐negative matrix factorization (NMF)‐based topic modeling with classifier‐based importance scoring to identify highly spatially informative genes, and fuses a local expression graph with a topic‐driven non‐local graph via consensus refinement and contrastive graph representation learning to jointly learn spatial and functional embeddings. Evaluated on 34 real‐world datasets, SAGE not only outperforms existing methods in clustering accuracy but also reveals functionally coherent regions and interpretable gene expression patterns. In case studies, SAGE reveals spatial heterogeneity associated with a pre‐malignant activation state in human breast cancer. Moreover, in zebrafish melanoma, it refines the tumor–muscle interface into transcriptionally distinct subdomains and uncovers shared vascular signatures between anatomically separate tissues. Together, these results demonstrate that SAGE can be used not only for accurate spatial domain delineation across diverse ST platforms, but also for dissecting microenvironmental niches and long‐range tissue interactions underlying disease progression.

Spatially resolved transcriptomics involves a set of emerging technologies that enable the transcriptomic profiling of tissues with the physical location of expressions. Although a variety of methods have been developed for data integration, most of them are for single-cell RNA-seq datasets without consideration of spatial information. Thus, methods that can integrate spatial transcriptomics data from multiple tissue slides, possibly from multiple individuals, are needed. Here, we present PRECAST, a data integration method for multiple spatial transcriptomics datasets with complex batch effects and/or biological effects between slides. PRECAST unifies spatial factor analysis simultaneously with spatial clustering and embedding alignment, while requiring only partially shared cell/domain clusters across datasets. Using both simulated and four real datasets, we show improved cell/domain detection with outstanding visualization, and the estimated aligned embeddings and cell/domain labels facilitate many downstream analyses. We demonstrate that PRECAST is computationally scalable and applicable to spatial transcriptomics datasets from different platforms.

We propose a method to leverage spatial transcriptomics and bioinformatic analysis to identify sub-clonal loss of heterozygosity that could otherwise be missed by standard bulk sequencing in glioblastoma tumors. Glioblastomas are highly invasive and aggressive brain tumors with a low five-year survival rate of less than 10%. Current treatment strategies are not sufficient for long term disease outcome, and innovative approaches to analyze glioblastoma are imperative for future improved options. Many factors contribute to difficulty of treatment, though one critical consideration is the diffuse nature and known intratumoral heterogeneity of this cancer. It is important to determine LOH heterogeneity, as this genomic alteration is associated with copy number variation and specific classifications of gliomas. Loss of heterozygosity is an irreversible deletion of DNA that is a common alteration in cancer. Application of spatial transcriptomics can increase our understanding of genomic complexity across a tissue, and unsupervised graph clustering can computationally separate spatial regions with similar gene expression patterns. Our method investigates spatial clusters individually at known heterozygous SNP sites to identify evidence for LOH with Bayesian inference and a hidden Markov model. Our analysis centers on the balance of reference (A) and non-reference (B) spatial transcriptomics read counts at SNPs of interest. B allele frequency is the number of reads aligned to the alternative allele divided by the total number of reads counted at the SNP. At a normal two-copy state at a heterozygous site, B allele frequency is ideally 50 percent. LOH can be identified as this value deviates toward 0 or 100, and copy number alterations impact this frequency. Our method utilizes Bayes factor values, or likelihood ratios between two models, using allele counts to provide support for either LOH or a heterozygous event. For our analysis, spatial transcriptomics BAMs are processed with a graph clustering algorithm and split into cluster-specific BAMs. We calculate allele count coverage for each cluster at pre-selected heterozygous SNP positions identified through germline exome sequencing. Bayes factor values are then calculated at each SNP, resulting in evidence for LOH or a heterozygous event. Chromosomal regions are segmented by a hidden Markov approach. Cumulative metrics for each segment are evaluated to determine a “state” label of heterozygous, LOH, or undefined. We tested our method on three WHO grade 4 IDH wild-type EGFRvIII positive fresh frozen glioblastoma tissue samples. Sub-clonal heterogeneity was identified in two samples and confirmed by bulk exome sequencing, bulk RNA sequencing, and analysis of initial clinical reports. Citation Format: Michelle G. Webb, Frances Chow, Carmel McCullough, Bohan Zhang, Kyle Hurth, John Carpten, Gabriel Zada, David W. Craig. Identification of sub-clonal loss of heterozygosity (LOH) in glioblastomas analyzed with spatial transcriptomics [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 3124.

stSCI: A multi-task learning framework for integrative analysis of single-cell and spatial transcriptomics data

Spatial transcriptomics is revolutionizing our understanding of tumor ecosystems, providing insights into the spatial dimensions of cell locations, their organization, and interactions within tissues. A key challenge in this field is integrating histological annotations with spatial domains identified through unsupervised clustering methods like SpaGCN. This integration is crucial for accurate spatial domain identification and further analyses. However, clustering results often vary based on parameters such as resolution, leading to inconsistencies with histological annotations. This variation necessitates optimized parameter selection to achieve the best clustering outcomes. We here introduce SpatialCS, a novel tool specifically designed to both visualize and quantify the alignment between histological annotations and the clustering results obtained from spatial transcriptomics data. SpatialCS employs a new index based on the asymmetric Wallace indices. This index penalizes spot pairs that are clustered together but do not share the same histology annotation type, with penalties scaled according to their transcriptional similarity. Additionally, we propose a purity index, derived by calculating the average transcriptional entropy of all clusters. This index serves as a measure of the homogeneity within each cluster. Furthermore, SpatialCS includes a unique visualization function. This function projects histological annotations onto the clustering results as outlines, making it easier to assess the consistency between histology annotations and clustering outcomes. We have applied SpatialCS to Visium spatial transcriptomics data generated on four colorectal cancer liver metastases, each accompanied by comprehensive spot-level histological annotations in both tumor and normal compartments. The histologic annotations include but not limited to are tumor cells, tumor cell cluster subtypes, tumor stroma, tumor in the vessels, portal tract, individual bile ducts, portal vein and liver cell plates and non-neoplastic stroma. The application of SpatialCS has proven its effectiveness in identifying the most consistent clustering results, which were further confirmed by experienced pathologists. Notably, the output of SpatialCS offers helpful guidance for improving histological annotations. Together, SpatialCS not only provides a quantitative approach to evaluate the consistency between histological annotations and unsupervised clustering analysis but also showcases its potential to significantly enhance histological annotations. Citation Format: Yanshuo Chu, Kyung Serk Cho, Isha Khanduri, Tieling Zhou, Jiahui Jiang, Yunhe Liu, Xinmiao Yan, Guangsheng Pei, Yibo Dai, Yang Liu, Ruiping Wang, Akshaya S Jadhav, Sharia Hernandez, Humam Kadara, Luisa Maren Solis Soto, Scott Kopetz, Dipen Maheshbhai Maru, Linghua Wang. SpatialCS: A tool to effectively integrate histological annotations and spatial transcriptomics analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 7417.

The identification of tumor cells is pivotal for understanding tumor heterogeneity and the tumor microenvironment. Recent advances in spatially resolved transcriptomics (SRT) have revolutionized the way that transcriptomic profiles are characterized and have enabled the simultaneous quantification of transcript locations in intact tissue samples. SRT is a promising alternative method to study gene expression patterns in spatial domains. Nevertheless, the precise detection of tumor regions within intact tissue remains a great challenge. A common strategy for identifying tumor cells is via tumor-specific marker gene expression signatures, which are highly dependent on marker accuracy. Another effective approach is through aneuploid copy number alterations, as most types of cancer exhibit copy number abnormalities. Here, we introduce a novel computational method, called TUSCAN (TUmor Segmentation and Classification ANalysis in spatial transcriptomics), which constructs a spatial copy number variation profile to improve the accuracy of tumor region identification. TUSCAN combines gene information from SRT data and hematoxylin-and-eosin-staining image to annotate tumor sections and other benign tissues. We benchmark the performance of TUSCAN and several existing methods through the application to multiple datasets from different SRT platforms. We demonstrate that TUSCAN can effectively delineate tumor regions, with improved accuracy compared to other approaches. Additionally, the output of TUSCAN provides interpretable clonal evolution inferences that may lead to novel insights into disease development and potential druggable targets.