Abstract Background: Spatial niches, often quantified through the co-occurrence or colocalization of various cell types, are a common measure of tissue organization in health and disease. However, a clear link is missing between how these spatial measures quantify tissue heterogeneity, how that heterogeneity relates to functional organization within the tissue, and how the functional organization impacts patient outcomes. Methods: The SpaceIQ™ multi-omics analysis platform addresses tissue heterogeneity through a formal analysis of spatial cell-cell communication from a mutual information perspective. This methodology allows for the virtual dissection of tissue into distinct tumor microenvironment (TME) programs. Inter-patient heterogeneity is then revealed by clustering these spatial TME programs based on their compositional fractions, a process independent of patient outcomes. Crucially, these spatial TME programs serve as highly effective "microdomains" that provide critical local context for cell-to-cell interactions and spatially modulated "network biology," demonstrating strong predictive power for patient outcomes. Results: We analyzed a publicly available 51-plex immunofluorescence based spatial proteomics data (CODEX platform) from checkpoint-treated cutaneous T-cell lymphoma patients, using spatial analysis based on microdomains and network biology. Our findings demonstrate that: (i) checkpoint expressions alone are poor predictors of patient response; (ii) spatial interactions between different cell types moderately improve prediction accuracy compared to multi-marker phenotypes; and (iii) microdomains significantly outperform non-spatial methods and checkpoint expression-based approaches in predicting response. Conclusions: Spatial analysis leveraging microdomains and network biology significantly enhance prediction accuracy compared to non-spatial or checkpoint expression-based methods. This advancement enables improved biomarker-driven patient selection and targeted therapy optimization. Citation Format: A. Burak Tosun, Raymond Yan, Brian Falkenstein, Filippo Pullara, S. Chakra Chennubhotla. Decoding tumor microenvironment heterogeneity through spatial microdomains and network biology to predict immunotherapy outcomes [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 7437.

Abstract Background: While pre-defined, or biased, phenotyping algorithms are a popular approach for analyzing spatial molecular data, offering ready biological interpretation (e.g., CD8+ T-cells, CD68+PD-L1+ macrophages) and reproducibility across labs with consistent threshold choices, they suffer from subjectivity, coarseness, and an inability to capture emergent biology. Conversely, unbiased phenotyping algorithms have the potential to address these limitations, but they currently lack straightforward biological interpretability. Methods: Our SpaceIQ™ multi-omics platform performs recursive cell typing (RCT) instead of standard hierarchical clustering and other graph-based algorithms for unbiased cell identification. RCT leverages the wide dynamic range (variance) in spatial proteomics, transcriptomics, and morphology data, driven by protein abundance and other technical factors, as biologically insightful. Unlike normalized standard methods, RCT allows markers with larger variances to drive early differentiation, with smaller-variance markers defining subsequent subpopulations. Results: We demonstrate the broad applicability of RCT using the SpaceIQ™ platform across three publicly available spatial datasets, including proteomics, transcriptomics, and brightfield pathology. To facilitate biological interpretation of the resulting unbiased cell populations, RCT approach: (i) identifies discriminatory biomarker signatures for annotating each RCT; (ii) calculates the probability of pre-defined phenotypes within any unbiased cell type; and (iii) generates a minimal marker panel that can approximate any given unbiased cell type with high probability. Conclusions: Unbiased cell typing, achieved through RCT, is critical in cancer research. This approach ensures the representation of all cell states—rare, abundant, positive, negative, and transitional—regardless of antigen expression. By not relying on extensive cell-type specific training, it is uniquely suited to capture the full spectrum of cellular heterogeneity. Citation Format: Filippo Pullara, Raymond Yan, Brian Falkenstein, A. Burak Tosun, S. Chakra Chennubhotla. Unbiased cell type identification and biological interpretation of spatial molecular data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6913.

Abstract Background: The rapid expansion of mIF imaging has increased the need for rigorous, standardized quality control. Variability introduced during sample prep, staining, and image acquisition can obscure biological signals and compromise downstream analysis. Common issues include batch effects that distort foreground-background contrast, acquisition errors causing over/underexposure or low contrast, and physical artifacts such as bubbles, debris, or deparaffinization defects. More difficult to detect is non-specific antibody binding, which can appear in unexpected structures or cell types. These challenges require a human-in-the-loop QC system capable of detecting, correcting, and documenting quality deviations across diverse tissues and platforms. Methods: Using a pan-tissue real-world mIF dataset, we evaluated the SpaceIQ™ QC pipeline across the most frequent image quality failure modes. QC begins with raw pixel data: a nuclear-derived tissue mask defines foreground/background regions for estimating and correcting batch effects. Channel-level parametric models detect blurring, saturation, bubbles, and other acquisition-related artifacts. Spatial modeling quantifies uneven illumination across tissue areas. A library of expected staining patterns is used to identify non-specific binding. Segmentation outputs are used to flag biologically impossible co-expression events (e.g., PanCK/CD45) as indicators of staining or acquisition issues. Each QC module generates a 0-3 score, from low quality/not-usable to no-artifacts, enabling a simple, interpretable quantitative assessment. All QC outputs are visualized in the SpaceIQ™ interface, which highlights artifact-containing ROIs and enables user confirmation or override. Results: The pipeline reliably detected and excluded regions affected by blurring, saturation, and uneven illumination. Applying standardized QC steps significantly altered cell counts and subtype distributions by correcting batch effects and removing regions affected by non-specific staining, demonstrating the importance of systematic QC before quantitative analysis. Conclusions: The SpaceIQ™ QC pipeline provides a platform- and tissue-agnostic framework for automated yet interpretable quality assessment of mIF imaging. By integrating artifact detection, batch correction, and biological validity checks, it improves accuracy, reproducibility, and user confidence, ensuring results reflect true biological signal rather than technical variability. Citation Format: Brian Falkenstein, A. Burak Tosun, Raymond Yan, S. Chakra Chennubhotla, Filippo Pullara. An end-to-end quality control pipeline for spatially resolved molecular imaging data in the multi-omic SpaceIQ™ platform [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6675.

Abstract Background: Ultra-high-plex multiplex immunofluorescence (mIF) imaging enables detailed characterization of the tumor microenvironment (TME), yet translating these rich datasets into clinically deployable, low-plex biomarkers remains a major barrier for precision immuno-oncology. Existing predictive models rely heavily on high-dimensional features or broad phenotypic panels, limiting scalability, interpretability, and routine pathology integration. A systematic framework is needed to compress spatially resolved molecular information into minimal, clinical-plex, yet highly informative signatures capable of predicting immunotherapy response. Methods: Our SpaceIQ™ platform is a multi-omic analysis tool that integrates spatial proteomics data with optimized feature selection algorithms to distill molecular signatures. We employ unbiased cell typing and microdomain discovery to identify a differentially expressed network of spatial interactions between these unbiased cell types, utilizing pointwise mutual information (PMI) analysis. Each interaction (either pairwise or higher-order cliques) within this network represents a potential spatial prognostic model capable of predicting patient response. A key component of our approach is the identification of a subset threshold cell population that is enriched for a given unbiased cell type using a low-plex panel. The final prognostic model for a differential clique combines a spatial proximity score with these low-plex marker intensities. Results: Analysis of ultra-high-plex (=51 markers) spatial data from trial specimens of checkpoint-treated cutaneous T-cell lymphoma patients demonstrated that tumor-immune and immune-immune interactions emerge as microdomains with minimal signatures of 6-8 markers and high prediction accuracies (AUC = 0.87, 95% CI (0.865-0.881)). Conclusions: The SpaceIQ platform enables the extraction of compact, clinically practical biomarker panels from ultra-high-plex mIF datasets without sacrificing predictive power. By linking spatial microdomain biology to sparse signature derivation, this framework supports scalable deployment of precision immunotherapy biomarkers and enhances patient-selection strategies in clinical practice. Citation Format: Raymond Yan, Brian Falkenstein, A. Burak Tosun, Filippo Pullara, S. Chakra Chennubhotla. Deriving high-fidelity, low-plex clinical signatures from ultra-high-plex spatial data for immunotherapy response prediction [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6666.

Abstract Background: Spatial imaging outputs continue to grow in scale and complexity. While brightfield IHC and H&E remain the qualitative gold standard for antibody-based assessment, mIF offers quantitative protein measurement on a single slide. However, challenges such as non-specific binding, imaging artifacts, and variability across sites and operators limit confidence in mIF reproducibility. A quantitative, robust method is needed to assess concordance between IHC and mIF stains. Methods: Using a pan-cancer dataset with a 4-plex mIF panel and matched IHC sections from consecutive slides, we first co-registered images into a shared coordinate space with Valis, applying global rigid and non-rigid transformations from feature matches. IHC stain channels were isolated via stain-matrix-based deconvolution. A tissue mask was generated on the mIF image using Otsu thresholding and morphology operations and then projected onto the IHC slide.Tissue was divided into tiles whose size accounted for section-to-section distance, registration error, and biological variability. Within each tile, random windows were sampled to perform two tests: (1) identify whether the tile contains high stain intensity and (2) determine whether the corresponding IHC and mIF tiles exhibit statistically concordant staining. This approach yields both a DICE score for high-stain region overlap and a stain concordance metric capturing agreement across high- and low-stain regions. Tile-level results and heatmaps are visualized in SpaceIQ™. Results: Concordance between mIF and IHC varied substantially across markers, with CD8 showing the highest and FoxP3 the lowest agreement, a trend consistent across samples. Concordance heatmaps also revealed strong spatial effects, with some tissue regions highly concordant and others clearly discordant. Expert visual review matched these quantitative findings. Conclusions: This segmentation-free framework identifies substantial marker- and region-specific variation in concordance between mIF and IHC staining. Because the method is marker-agnostic and compensates for registration error and inter-section biological differences, it provides a generalized, quantitative approach for evaluating agreement between paired mIF and IHC slides across platforms. Citation Format: Brian Falkenstein, Raymond Yan, A. Burak Tosun, S. Chakra Chennubhotla, Filippo Pullara. Robust segmentation-free stain quality concordance metrics in the SpaceIQ™ multi-omic analysis platform [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 715.

Spatial biology has demonstrated that therapeutic responses are shaped by the organization and abundance of tumor, immune, and stromal cells within tissue, including specific microenvironments. Spatial omics, multiplex imaging, and computational pathology captures these architectures with increasing detail, yet most spatial discoveries remain difficult to translate into clinical trials, deployable biomarkers and precision therapeutic selection. The central barrier in the field has evolved from generating data to the development of mature platforms that translate complex tissue maps into analytically reproducible, clinically deployable, and decision-ready spatial outputs for precision oncology and drug development. Here, we propose a framework for converting spatial biology into decision-oriented clinical platforms built on quantitative spatial quality control, biologically grounded abstractions, interpretable modeling, and clinical compression layers.

Abstract Background: Idiopathic pulmonary fibrosis (IPF) is characterized by progressive alveolar injury and extensive tissue remodeling that generate pronounced spatial heterogeneity across affected lung regions. Distinct pathological programs often coexist within adjacent microenvironments, reflecting cellular interactions and molecular circuits that drive disease progression. Although histologic assessment of H&E-stained biopsies remains central to clinical diagnosis, current evaluations lack quantitative measures that connect visual pathology to underlying biology. Integrating spatial transcriptomics with quantitative histologic features enables higher-resolution characterization of heterogeneous disease regions and provides molecular context essential for accurate diagnostic interpretation. Methods: We applied a multi-omic analytic workflow using the SpaceIQ™ platform to the publicly available Xenium H&E and spatial transcriptomics dataset reported by Vannan et al., Nat. Genet. 2025. Unbiased cell typing was performed on H&E images and spatially aligned to cell-resolved gene expression. Spatial microdomains were derived from differential organization of inferred cell populations in IPF versus healthy tissue. Gene-feature associations were computed to identify molecular markers linked to pathology-associated microdomains. Candidate biomarkers were evaluated in an independent Visium dataset from Mayr et al., Sci. Adv. 2023 to assess reproducibility across platforms and cohorts. Results: Unbiased H&E-derived cell typing revealed finer structural organization within pathologist-annotated regions and captured subregional distinctions in fibrotic and non-fibrotic compartments. Spatial microdomains identified from cell-type arrangements distinguished IPF-specific architectural patterns and yielded gene signatures associated with epithelial dysregulation, inflammatory macrophages, and extracellular matrix remodeling. These signatures showed consistent enrichment in corresponding molecular niches within the validation cohort. Cross-cohort mapping further demonstrated reproducible cell-type differences and implicated conserved pathways involved in epithelial stress responses, fibroblast activation, and tissue remodeling. Conclusions: Integrating quantitative histologic features with spatial transcriptomics provides a robust framework for linking visual pathology to molecular mechanisms in IPF. Histology-derived microdomains reveal reproducible biological programs across independent cohorts and support the identification of clinically relevant biomarkers. This multi-omic approach enables more objective, biologically grounded interpretation of heterogeneous fibrotic regions and has the potential to improve diagnostic evaluation and patient stratification in IPF. Citation Format: Raymond Yan, Brian Falkenstein, A. Burak Tosun, Filippo Pullara, S. Chakra Chennubhotla. Spatial microdomains from histology reveal multi-omic biomarkers for enhanced idiopathic pulmonary fibrosis diagnosis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1487.

The emergence of single-cell spatial omics platforms generating high-plex proteomic and transcriptomic measurements sets the foundation for accurate downstream biologically interpretable analyses. This includes biased and unbiased approaches for cell typing, cell-cell interactions and the discovery of microdomains, complex multicellular microenvironments, underlying disease progression. There is a lack of a computational framework for optimizing unbiased approaches with biological priors for unraveling a deeper understanding of disease mechanisms. This requires innovative integrational methods that assign confidence and robustness to data-driven biological hypotheses, e.g., presence of transition cells and other rare cell types, emergent microdomains and spatially modulated network biology. We present a Bayesian computational approach, SpaceIQ™, that not only elucidates potential novel hypotheses driving disease mechanisms, but also cross-references existing working hypotheses from alternative methods. SpaceIQ is agnostic to imaging platforms with the ability to ingest any combinatorial forms of spatial data (e.g., transmitted light, proteomics and transcriptomics). To illustrate our approach, we compare the results from SpaceIQ to the interpretative analyses presented in He et al. Nat Biotech 2022, a publicly available 960-plex RNA data from CosMx platform on NSCLC samples. We aim to provide additional mechanistic insights underlying Stage II/III progression that can augment the biological annotations reported. We have identified 16 unbiased cell types with probabilistic representation of a priori phenotypes annotated as myeloid, lymphocytes, endothelial, epithelial/tumor, and fibroblast. IFI27, SOX4, MALAT1, TYK2, CD74, HLADRB1, and COL3A1 were among the highly-expressed discriminatory genes among cell types. 28 significant spatial interactions (p<0.1) were found resulting in 7 microdomains. Comparison of microdomains to niche neighborhoods defined in He et al. shows that macrophages in the stroma and TLS play a role in tumor progression. Ligand-Receptor analysis on microdomains identified IL-2 signaling, GPCR ligand binding, TNF activity, and TGF regulation to be significant cell-cell communication channels in NSCLC progression (p<1e-04). We proposed a Bayesian integration framework to further the biological interpretations of the spatially-resolved high-plex single-cell CosMx dataset on NSCLC patients presented in He et al. We’ve found unbiased cell types linked to known cell phenotypes in addition to potentially novel gene signatures. These unbiased cell types formulate microdomains with identified key signaling events that are associated with cancer progression. Brian Falkenstein, Raymond Yan, Shannon Quinn, Akif Burak Tosun, S. Chakra Chennubhotla, Arutha Kulasinghe, Filippo Pullara. A Bayesian framework for unraveling disease biology from spatially resolved single-cell omics datasets by combining unbiased approaches with biological priors for cell-based microdomain discovery [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2485.

Colorectal cancer is the second most fatal cancer worldwide, accounting for 9% of all reported cancer deaths, and is the number one cause of death among people under the age of 50. Rapid technical advances in the field of spatial biology are allowing us to gain new insight into colorectal cancer biology and patient heterogeneity, which can help to guide the development of immunotherapies. Using a combination of modalities such as sequential multiplex immunofluorescence (mIF) and spatial transcriptomics (STx) on a colon cancer sample we determined the spatial distribution, phenotype, function, and a cytokine and chemokine profile within the colon cancer tissue. 5 um FFPE sections were prepared from a colon cancer sample exhibiting moderate to poorly differentiated adenocarcinoma of T3, N0 staging. mIF was performed using Lunaphore COMET with a 20-antibody panel, followed by H&E staining and imaging. An adjacent section was used for STx with the 10x Genomics Visium CytAssist platform. Data was analyzed with Lunaphore Horizon®, 10x Genomics Loupe Browser software and PredxBio SpaceIQ™ unbiased cell typing, microdomain discovery and network biology platform that co-analyzes mIF and STx datasets. The tumor showed an aberrant growth pattern and a highly proliferated crypt region. Three distinct microdomains in the tumor tissue were discovered: one with high levels of PD-L1+ cells, a second with high levels of NK cells and a third with low levels of infiltrating cells in the lumen and crypt. There was significant presence of VISTA+ and IDO1+ cells throughout the lumen. These cells interacted with CD4+ and CD8+ cells which were found to be present in both lumen and crypt regions. PD-L1 was prominent in the cells found in both lumen and crypt. Finally, NK cells were abundant throughout the lumen and in regions adjacent to the crypt. Using multimodal spatial imaging and spatial co-analysis of mIF and STx data, we were able to elucidate relationships between different cell types in a complex tumor tissue and assign context to their activities. Despite high levels of NK cells, indicative of a possible active anti-tumor immune response, the tumor has developed an immune evasion mechanism due to the presence of cells expressing high levels of the anti-tumor immune response markers such as PD-L1, IDO1, and VISTA. Based on these results, therapies targeting more than one immune checkpoint protein may yield improved response due to an abundance of the existing NK cells. The transcriptomic profiling reveals a complex relationship between various cytokines secreted by different cell populations which alter the landscape of the tumor microenvironment, facilitating the emergence of an immunosuppressive phenotype. Gaurav Joshi, Nicholas Sciascia, Michael Yang, Brian Falkenstein, Raymond Yan, Shannon Quinn, Brenna Fearey, Junya Yoshioka, Arif Burak Tosun, S. Chakra Chennubhotla, Filippo Pullara, Fumiki Yanagawa. Multimodal spatial analysis of colon cancer tissue reveals emergence of an immunosuppressive tumor maintenance mechanism [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 5317.

Histopathological images, which are digitized images of human or animal tissue, contain insights into disease state. Typically, a pathologist will look at a slide under a microscope to make decisions about prognosis and treatment. Due to the high complexity of the data, applying automatic image analysis is challenging. Often, human intervention in the form of manual annotation or quality control (QC) is required. Additionally, the data itself varies considerably in available features, size, and shape. Thus, a streamlined and interactive approach is a necessary part of any digital pathology pipeline. We present PredX-Tools, a suite of simple and easy to use python GUI applications which facilitate analysis of histopathological images and provide a no-code platform for data scientists and researchers to perform analysis on raw and transformed data.

Ca2+/calmodulin-dependent protein kinase II α (CaMKIIα) signaling in the brain plays a critical role in regulating neuronal Ca2+ homeostasis. Its dysfunctional activity is associated with various neurological and neurodegenerative disorders, including Parkinson's disease (PD). Using computational modeling analysis, we predicted that, two essential cysteine residues contained in CaMKIIα, Cys30 and Cys289, may undergo redox modifications impacting the proper functioning of the CaMKIIα docking site for Ca2+/CaM, thus impeding the formation of the CaMKIIα:Ca2+/CaM complex, essential for a proper modulation of CaMKIIα kinase activity. Our subsequent in vitro investigations confirmed the computational predictions, specifically implicating Cys30 and Cys289 residues in impairing CaMKIIα:Ca2+/CaM interaction. We observed CaMKIIα:Ca2+/CaM complex disruption in dopamine (DA) nigrostriatal neurons of post-mortem Parkinson's disease (PD) patients' specimens, addressing the high relevance of this event in the disease. CaMKIIα:Ca2+/CaM complex disruption was also observed in both in vitro and in vivo rotenone models of PD, where this phenomenon was associated with CaMKIIα kinase hyperactivity. Moreover, we observed that, NADPH oxidase 2 (NOX2), a major enzymatic generator of superoxide anion (O2●-) and hydrogen peroxide (H2O2) in the brain with implications in PD pathogenesis, is responsible for CaMKIIα:Ca2+/CaM complex disruption associated to a stable Ca2+CAM-independent CaMKIIα kinase activity and intracellular Ca2+ accumulation. The present study highlights the importance of oxidative stress, in disturbing the delicate balance of CaMKIIα signaling in calcium dysregulation, offering novel insights into PD pathogenesis.

Altered Ca2+/calmodulin-dependent protein kinase II (CaMKII) signaling in nigrostriatal dopamine (DA) neurons is considered a major cause of Ca2+ signaling dysregulation in neurodegenerative disorders, including Parkinson’s disease (PD). Increasing number of evidence addresses oxidative stress - a key pathogenic event in PD – as inducer of a sustained Ca2+/CaM-independent CaMKII activity. Here, applying computational modeling, we identified two critical Cysteine (Cys) residues that, in a redox dependent manner, may impact the proper function of the CaMKII docking site for Ca2+/CaM, leading to the disruption of the CaMKII:Ca2+/CaM complex. In vitro investigations confirmed the computational prediction and revealed an implication of Cys30 and Cys289 residues in the loss of CaMKII:Ca2+/CaM interaction. Noteworthy, this phenomenon was also observed in Dopamine (DA) nigrostriatal neurons from post-mortem PD brain specimens and in in vitro and in vivo rotenone models of PD, revealing a functional link between CaMKII:Ca2+/CaM complex disruption and CaMKII kinase hyperactivity. The present study, thus. provides novel insights into a redox-mediated dysfunction of CaMKII:Ca2+/CaM complex which may contribute to the PD pathogenesis.

Background: The spatial intratumor heterogeneity (ITH) is widely acknowledged as driving therapeutic response and providing fuel for drug resistance. Currently, patient selection for immunotherapy is driven mostly by PD-1/PD-L1 based IHC tests and mutational analysis. These oversimplified approaches fail to predict the risk of recurrence, therapeutic response and drug resistance with high accuracy. We hypothesize that functional responses of heterogeneous non-random spatial arrangements of tumor, stromal and immune cells in the tumor microenvironment are determined by distinct combinations of their internal states and spatial interactions within neighborhoods. The interactions generate distinct cell-cell communication networks forming functional spatial configurations of cells, termed microdomains, as organizational units of spatial ITH. Deriving the spatial networks within each microdomain with unbiased spatial analytics and the underlying network biology through explainable AI, is key to understanding tumor initiation, tumor progression, and response to therapy. Problem: There has been an explosion of spatial imaging technologies using immunofluorescence and/or mass spectrometry for intact tissues measuring protein expressions, DNA and RNA probes. To extract high-value knowledge from these multiplexed datasets, a key first step is to segment cells accurately. Despite decades of research, this step remains elusive due to imaging artifacts arising from issues with tile stitching, incorrect registration, signal oversaturation, non-uniform illumination, poor or high-background nuclei signal, defocus, varying cell and nuclear shapes, fluorescence emission efficiency variation, overlapping and/or superimposed nuclei and low resolution (e.g., mass cytometry). Those artifacts may lead to incorrect cell phenotypes, incomplete cell phenotype atlases, and missing rare cell, fusion cell or transition cell types. Solution: We present TumorMapr™, a segmentation-free, unbiased spatial analytics and explainable AI platform that extracts information and creates knowledge from patient disease pathology samples imaged with any technology including fluorescence, mass spectrometry etc. Results: We apply TumorMapr on hyperplexed immunofluorescence dataset of colorectal cancer and hyperplexed imaging mass cytometry dataset of triple negative breast cancer to discover functional cell prototypes involving spatial collections of neighboring pixels highly predictive of disease progression and response to therapy, tumor promoting and tumor restraining microdomains and microdomain-specific network biology predictive of disease outcomes. We further compare these results with segmentation-based approaches to showcase the discovery of novel cell prototypes. Citation Format: Filippo Pullara, Brian Falkenstein, Bruce Campbell, Samantha Panakkal, Akif Burak Tosun, Jeffrey Fine, S. Chakra Chennubhotla. Segmentation-free analysis of multiplexed images with unbiased spatial analytics and explainable AI for predicting disease outcomes. [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 5445.

Background: The current computational analyses of multi to hyperplexed fluorescence and/or mass spectrometry image datasets from patient pathology samples are not powerful enough to extract the maximum amount of information or to create the detailed knowledge that is required to advance precision medicine in pathology, including the development of personalized therapeutic strategies, identification of potential novel targets for drug discovery, selection of optimal patient cohorts for clinical trials, and improvement of the predictive power of prognostics/diagnostics. Methods: TumorMapr harnesses the computational power of proprietary, unbiased spatial analytics, spatial systems pathology, and explainable artificial intelligence (xAI) to extract information and to create knowledge from patient primary disease pathology samples imaged on any of the existing fluorescence and/or mass spectrometry imaging platforms. Results: To demonstrate the generalizability and utility of the TumorMapr platform, we apply it on two different datasets: hyperplexed immunofluorescence-based colorectal cancer data (51 biomarkers, 431 patients) and imaging mass cytometry-based breast cancer data (35 biomarkers, 281 patients). The TumorMapr platform (i) unlike the biased intensity thresholding approaches, the unbiased and automated functional cell phenotyping discovers a continuum of cell types and cell states, including transitional, multi-transitional cell states and fusion cell types that are critical for disease progression; (ii) derives microdomains with tumor promoting and tumor suppressing properties that are highly predictive of disease progression and response to therapy; (iii) spatial systems pathology analysis taps into the current network biology knowledge databases to derive pathway interactions and signaling networks, identify novel biomarkers and potential molecular targets and drugs, in the spatial context of each microdomain; (iv) xAI application guide, for example, in the case of predicting 5-year risk of recurrence in CRC patients, provides explanations in the form of microdomain-specific networks that are driving disease progression. Using the TumorMapr pipeline we created a prognostic test that shows a vastly superior performance over current approaches in predicting 5-yr risk of recurrence in CRC patients. Further, the TumorMapr platform enables building a rich outcome-specific library of microdomains to directly apply on prospective tissue samples for a companion diagnostic test that predicts disease outcomes. Citation Format: Samantha Panakkal, Brian Falkenstein, Akif Burak Tosun, Bruce Campbell, Michael Becich, Jeffrey Fine, D. Lansing Taylor, S. Chakra Chennubhotla, Filippo Pullara. TumorMapr™ analytical software platform: Unbiased spatial analytics and explainable AI (xAI) platform for generating data, extracting information, and creating knowledge from multi to hyperplexed fluorescence and/or mass spectrometry image datasets [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 454.