This study systematically reviews and evaluates published research on machine learning models that integrate histopathology whole slide images and high-throughput -omic data to predict overall survival in cancer. A comprehensive search of PubMed, EMBASE, and Cochrane CENTRAL was conducted through August 12, 2024, with citation screening for additional studies. Eligible studies applied machine learning or deep learning methods to multimodal data combining pathology images and -omics. Data extraction followed the CHARMS checklist, and risk of bias was assessed using the PROBAST + AI tool. Narrative synthesis was conducted in line with PRISMA 2020 guidelines. Forty-eight studies published since 2017 met inclusion criteria, spanning 19 cancer types. All relied on The Cancer Genome Atlas dataset. Modelling approaches included regularised Cox regression (n = 4), classical machine learning (n = 13), and deep learning (n = 31). Reported concordance indices ranged from 0.550 to 0.857, with most multimodal models outperforming unimodal counterparts. However, all studies were assessed as having high or unclear risk of bias—most often due to limited external validation, insufficient reporting, and minimal assessment of clinical utility. This review highlights a rapidly evolving yet methodologically underdeveloped field. While model performance is promising, improvements in data standardisation, reporting practices, and real-world contextualisation are critical for clinical translation. This work was funded by the National Pathology Imaging Cooperative (NPIC), supported by UK Research and Innovation (Project no. 104687).

Given the rapid resistance of Plasmodium falciparum to antimalarial drugs, there is a continual need for new treatments. A genome-scale metabolic (GSM) model was developed with integrated metabolomics and constraint-based, experimental flux-balance data to predict genes essential for P. falciparum growth as drug targets. We selected the highly ranked P. falciparum UMP-CMP kinase (UCK) to test its necessity and the ability to inhibit parasite growth in the presence of inhibitors. Conditional deletion mutants using the DiCre recombinase system, generated by CRISPR-Cas genome editing, exhibited defective asexual growth and stage-specific developmental arrest. Based on in silico and in vitro screening, inhibitors were identified that are selective for P. falciparum UCK and exhibit antiparasitic activity. This study, for the first time, shows assertions from a GSM model identifying novel, validated "druggable" targets. These findings show a role for GSM models in antimalarial drug discovery and identify P. falciparum UCK as a novel, valid malaria drug target.

Multimodal machine learning integrating histopathology and molecular data shows promise for cancer prognostication. We systematically reviewed studies combining whole slide images (WSIs) and high-throughput omics to predict overall survival. Searches of EMBASE, PubMed, and Cochrane CENTRAL (12/08/2024), plus citation screening, identified eligible studies. Data extraction used CHARMS; bias was assessed with PROBAST+AI; synthesis followed SWiM and PRISMA 2020. Protocol: PROSPERO (CRD42024594745). Forty-eight studies (all since 2017) across 19 cancer types met criteria; all used The Cancer Genome Atlas. Approaches included regularised Cox regression (n=4), classical ML (n=13), and deep learning (n=31). Reported c-indices ranged 0.550-0.857; multimodal models typically outperformed unimodal ones. However, all studies showed unclear/high bias, limited external validation, and little focus on clinical utility. Multimodal WSI-omics survival prediction is a fast-growing field with promising results but needs improved methodological rigor, broader datasets, and clinical evaluation. Funded by NPIC, Leeds Teaching Hospitals NHS Trust, UK (Project 104687), supported by UKRI Industrial Strategy Challenge Fund.

Extrachromosomal DNA amplification is associated with poor cancer prognoses 1 . Large numbers of excised signal circles (ESCs) are produced as by-products of antigen receptor rearrangement during V(D)J recombination 2,3 . However, current dogma states that ESCs are progressively lost through cell division 4 . Here we show that ESCs replicate and persist through many cell generations and share many properties in common with circular extrachromosomal DNAs. Increased ESC copy numbers at diagnosis of B cell precursor acute lymphoblastic leukaemia were highly correlated with subsequent relapse. By taking advantage of the matching recombination footprint that is formed upon the generation of each ESC, we measured ESC persistence and replication and found increased ESC replication in patients who later relapsed. This increased replication is controlled by cell-intrinsic factors and corresponds to increased expression of DNA replication- and repair-associated genes. Consistent with high ESC levels having a role in disease progression, the number of mutations typical of those caused by the V(D)J recombinase–ESC complex was significantly increased at diagnosis in patients who later relapsed. The number of such mutations in genes associated with relapse increased between diagnosis and relapse, and corresponded to clonal expansion of cells with high ESC copy numbers. These data demonstrate that the by-product of V(D)J recombination, when increased in abundance, potently associates with the V(D)J recombinase to cause adverse disease outcomes.

Motivation: Networks underlie the generation and interpretation of many biological datasets: gene networks shed light on the regulatory structure of the genome, and cell networks can capture structure of the tumor micro-environment. However, most methods that learn such networks make the faulty 'independence assumption'; to learn the gene network, they assume that no cell network exists. 'Multi-axis' methods, which do not make this assumption, fail to scale beyond a few thousand cells or genes. This limits their applicability to only the smallest datasets. Results: We develop a multi-axis method capable of processing million-cell datasets within minutes. This was previously impossible, and unlocks the use of such methods on modern scRNA-seq datasets, as well as more complex datasets. We show that our method yields novel biological insights from real single-cell data, and compares favorably to the existing hdWGCNA methodology. In particular, it identifies long non-coding RNA genes that potentially have a regulatory or functional role in neuronal development. Availability and implementation: Our methodology is available as a Python package GmGM on PyPI (https://pypi.org/project/GmGM/0.5.3/). The code for all experiments performed in this paper is available on GitHub (https://github.com/BaileyAndrew/GmGM-Bioinformatics). Contact: sceba@leeds.ac.uk Supplementary information: Our proofs, and some additional experiments, are available in the supplementary material. Keywords: gaussian graphical models, multi-axis models, transcriptomics, multi-omics, scalability

MYC translocation occurs in 8-14% of diffuse large B-cell lymphoma (DLBCL), and may concur with BCL2 and/or BCL6 translocation, known as double-hit (DH) or triple-hit (TH). DLBCL-MYC/BCL2-DH/TH are largely germinal centre B-cell like subtype, but show variable clinical outcome, with IG::MYC fusion significantly associated with inferior survival. While DLBCL-MYC/BCL6-DH are variable in their cell-of-origin subtypes and clinical outcome. Intriguingly, only 40-50% of DLBCL with MYC translocation show high MYC protein expression (>70%). We studied 186 DLBCLs with MYC translocation including 32 MYC/BCL2/BCL6-TH, 75 MYC/BCL2-DH and 26 MYC/BCL6-DH. FISH revealed a MYC/BCL6 fusion in 59% of DLBCL-MYC/BCL2/BCL6-TH and 27% of DLBCL-MYC/BCL6-DH. Targeted NGS showed a similar mutation profile and LymphGen genetic subtype between DLBCL-MYC/BCL2/BCL6-TH and DLBCL-MYC/BCL2-DH, but variable LymphGen subtypes among DLBCL-MYC/BCL6-DH. MYC protein expression is uniformly high in DLBCL with IG::MYC, but variable in those with non-IG::MYC including MYC/BCL6-fusion. Translocation breakpoint analyses of 8 cases by TLC-based NGS showed no obvious genomic configuration that enables MYC transactivation in 3 of the 4 cases with non-IG::MYC, while a typical promoter substitution or IGH super enhancer juxtaposition in the remaining cases. The findings potentially explain variable MYC expression in DLBCL with MYC translocation, and also bear practical implications in its routine assessment.

A framework for the rational selection of a minimal suite of nondegenerate developability assays (DAs) that maximize insight into candidate developability or storage stability is lacking. To address this, we subjected nine formulation:mAbs to 12 mechanistically distinct DAs together with measurement of their accelerated and long-term storage stability. We show that it is possible to identify a reduced set of key variables from this suite of DAs by using orthogonal statistical methods. We exemplify our approach by predicting the rank formulation:mAb degradation rate at 25 °C (determined over 6 months) using just five DAs that can be measured in less than 1 day, spanning a range of physicochemical features. Implementing such approaches focuses on resources, thus increasing sustainability and decreasing development costs.

BackgroundBrain metastases (BrM) affect up to 60% of patients with metastatic melanoma and are associated with poor prognosis. While combined immune checkpoint blockade of programmed death-1 (PD-1) and cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) demonstrates intracranial efficacy in a proportion of patients with melanoma, the responses are rarely durable, particularly in patients with symptomatic BrM. The brain is an immune-specialized organ and immune responses are regulated differently to the periphery.MethodsUsing our previously established two-site model of melanoma BrM with concomitant intracranial and extracranial tumors, in which clinically observed efficacy of the combined PD-1/CTLA-4 (PC) blockade can be reproduced, we here explored the role of natural killer (NK) cells in BrM, using functional studies, immunophenotyping and molecular profiling.ResultsWe demonstrate that NK cells are required for the intracranial efficacy of PC blockade. While both perforin and interferon gamma were necessary for the PC blockade-dependent control of intracranial tumor growth, NK cells isolated from intracranial tumors demonstrated only a limited cancer cell killing ability, and PC blockade did not alter the abundance of NK cells within tumors. However, the depletion of NK cells in PC blockade-treated mice led to tumor molecular profiles reminiscent of those observed in intracranial tumors that failed to respond to therapy. Furthermore, the depletion of NK cells resulted in a strikingly reduced abundance of CD8+ T cells within intracranial tumors, while the abundance of other immune cell populations including CD4+ T cells, macrophages and microglia remained unaltered. Adoptive T cell transfer experiments demonstrated that PC blockade-induced trafficking of CD8+ T cells to intracranial tumors was chemokine-dependent. In line with this, PC blockade enhanced intratumoral expression of several T cell-attracting chemokines and we observed high expression levels of cognate chemokine receptors on BrM-infiltrating CD8+ T cells in mice, as well as in human BrM. Importantly, the depletion of NK cells strikingly reduced the intratumoral expression levels of T cell attracting chemokines and vascular T cell entry receptors that were upregulated following PC blockade.ConclusionOur data demonstrate that NK cells underpin the efficacy of PC blockade in BrM by orchestrating the "responder" molecular profile in tumors, and by controlling the intratumoral abundance of CD8+ T cells through regulation of multiple key molecular mediators of T cell trafficking.

This paper introduces the Gaussian multi-Graphical Model, a model to construct sparse graph representations of matrix- and tensor-variate data. We generalize prior work in this area by simultaneously learning this representation across several tensors that share axes, which is necessary to allow the analysis of multimodal datasets such as those encountered in multi-omics. Our algorithm uses only a single eigendecomposition per axis, achieving an order of magnitude speedup over prior work in the ungeneralized case. This allows the use of our methodology on large multi-modal datasets such as single-cell multi-omics data, which was challenging with previous approaches. We validate our model on synthetic data and five real-world datasets.

Background Glioblastoma (GBM) brain tumors lacking IDH1 mutations (IDHwt) have the worst prognosis of all brain neoplasms. Patients receive surgery and chemoradiotherapy but tumors almost always fatally recur. Results Using RNA sequencing data from 107 pairs of pre- and post-standard treatment locally recurrent IDHwt GBM tumors, we identify two responder subtypes based on longitudinal changes in gene expression. In two thirds of patients, a specific subset of genes is upregulated from primary to recurrence (Up responders), and in one third, the same genes are downregulated (Down responders), specifically in neoplastic cells. Characterization of the responder subtypes indicates subtype-specific adaptive treatment resistance mechanisms that are associated with distinct changes in the tumor microenvironment. In Up responders, recurrent tumors are enriched in quiescent proneural GBM stem cells and differentiated neoplastic cells, with increased interaction with the surrounding normal brain and neurotransmitter signaling, whereas Down responders commonly undergo mesenchymal transition. ChIP-sequencing data from longitudinal GBM tumors suggests that the observed transcriptional reprogramming could be driven by Polycomb-based chromatin remodeling rather than DNA methylation. Conclusions We show that the responder subtype is cancer-cell intrinsic, recapitulated in in vitro GBM cell models, and influenced by the presence of the tumor microenvironment. Stratifying GBM tumors by responder subtype may lead to more effective treatment.

Summary: Background: Knowledge of post-myocardial infarction (MI) disease risk to date is limited—yet the number of survivors of MI has increased dramatically in recent decades. We investigated temporally ordered sequences of all conditions following MI in nationwide electronic health record data through the application of process mining. Methods: We conducted a national retrospective cohort study of all hospitalisations (145,670,448 episodes; 34,083,204 individuals) admitted to NHS hospitals in England (1st January 2008–31st January 2017, final follow-up 27th March 2017). Through process mining, we identified trajectories of all major disease diagnoses following MI and compared their relative risk (RR) and all-cause mortality hazard ratios (HR) to a risk-set matched non-MI control cohort using Cox proportional hazards and flexible parametric survival models. Findings: Among a total of 375,669 MI patients (130,758 females; 34.8%) and 1,878,345 matched non-MI patients (653,790 females; 34.8%), we identified 28,799 unique disease trajectories. The accrual of multiple circulatory diagnoses was more common amongst MI patients (RR 4.32, 95% CI 3.96–4.72) and conferred an increased risk of death (HR 1.32, 1.13–1.53) compared with matched controls. Trajectories featuring neuro-psychiatric diagnoses (including anxiety and depression) following circulatory disorders were markedly more common and had increased mortality post MI (HR ranging from 1.11 to 1.73) compared with non-MI individuals. Interpretation: These results provide an opportunity for early intervention targets for survivors of MI—such as increased focus on the psychological and behavioural pathways—to mitigate ongoing adverse disease trajectories, multimorbidity, and premature mortality. Funding: British Heart Foundation; Alan Turing Institute.

Despite the effectiveness of immuno-chemotherapy, 40% of patients with diffuse large B-cell lymphoma (DLBCL) experience relapse or refractory disease. Longitudinal studies have previously focused on the mutational landscape of relapse but fell short of providing a consistent relapse-specific genetic signature. In our study, we have focused attention on the changes in GEP accompanying DLBCL relapse using archival paired diagnostic/relapse specimens from 38 de novo patients with DLBCL. COO remained stable from diagnosis to relapse in 80% of patients, with only a single patient showing COO switching from activated B-cell-like (ABC) to germinal center B-cell-like (GCB). Analysis of the transcriptomic changes that occur following relapse suggest ABC and GCB relapses are mediated via different mechanisms. We developed a 30-gene discriminator for ABC-DLBCLs derived from relapse-associated genes that defined clinically distinct high-and low-risk subgroups in ABC- DLBCLs at diagnosis in datasets comprising both population-based and clinical trial cohorts. This signature also identified a population of <60-year-old patients with superior PFS and OS treated with ibrutinib-R-CHOP as part of the PHOENIX trial. Altogether this new signature adds to the existing toolkit of putative genetic predictors now available in DLBCL that can be readily assessed as part of prospective clinical trials.

A limited understanding of the pathology underlying chronic wounds has hindered the development of effective diagnostic markers and pharmaceutical interventions. This study aimed to elucidate the molecular composition of various common chronic ulcer types to facilitate drug discovery strategies. We conducted a comprehensive analysis of leg ulcers (LUs), encompassing venous and arterial ulcers, foot ulcers (FUs), pressure ulcers (PUs), and compared them with surgical wound healing complications (WHCs). To explore the pathophysiological mechanisms and identify similarities or differences within wounds, we dissected wounds into distinct subregions, including the wound bed, border, and peri-wound areas, and compared them against intact skin. By correlating histopathology, RNA sequencing (RNA-Seq), and immunohistochemistry (IHC), we identified unique genes, pathways, and cell type abundance patterns in each wound type and subregion. These correlations aim to aid clinicians in selecting targeted treatment options and informing the design of future preclinical and clinical studies in wound healing. Notably, specific genes, such as PITX1 and UPP1, exhibited exclusive upregulation in LUs and FUs, potentially offering significant benefits to specialists in limb preservation and clinical treatment decisions. In contrast, comparisons between different wound subregions, regardless of wound type, revealed distinct expression profiles. The pleiotropic chemokine-like ligand GPR15L (C10orf99) and transmembrane serine proteases TMPRSS11A/D were significantly upregulated in wound border subregions. Interestingly, WHCs exhibited a nearly identical transcriptome to PUs, indicating clinical relevance. Histological examination revealed blood vessel occlusions with impaired angiogenesis in chronic wounds, alongside elevated expression of genes and immunoreactive markers related to blood vessel and lymphatic epithelial cells in wound bed subregions. Additionally, inflammatory and epithelial markers indicated heightened inflammatory responses in wound bed and border subregions and reduced wound bed epithelialization. In summary, chronic wounds from diverse anatomical sites share common aspects of wound pathophysiology but also exhibit distinct molecular differences. These unique molecular characteristics present promising opportunities for drug discovery and treatment, particularly for patients suffering from chronic wounds. The identified diagnostic markers hold the potential to enhance preclinical and clinical trials in the field of wound healing.

This paper introduces the Gaussian multi-Graphical Model, a model to construct sparse graph representations of matrix- and tensor-variate data. We generalize prior work in this area by simultaneously learning this representation across several tensors that share axes, which is necessary to allow the analysis of multimodal datasets such as those encountered in multi-omics. Our algorithm uses only a single eigendecomposition per axis, achieving an order of magnitude speedup over prior work in the ungeneralized case. This allows the use of our methodology on large multi-modal datasets such as single-cell multi-omics data, which was challenging with previous approaches. We validate our model on synthetic data and five real-world datasets.

Models that more accurately reflect Ewing sarcoma (ES) will enable the prioritization of novel targeted agents from bench to clinic. To date, development of an ES mouse model have been unsuccessful1 and engraftment of human ES in mice is variable2. Coupled with a moral obligation to minimize the use of mammals, we have established and characterized models using patient-derived ES cultures in vitro and in larval zebrafish. The transcriptomes of ES cell lines and patient-derived ES cultures3 were analyzed and compared using total RNAseq followed by DESeq2 and Gene Set Enrichment analysis. Cells in 2D, as 3D spheroids, in combination with cells of the tumor microenvironment and in larval zebrafish (Danio rerio) were characterized. Six cell lines clustered independently from patient-derived ES cultures (n=20), reflecting differential expression of 15,709 RNAs (p<0.01). Interrogation of the differentially expressed genes using Reactome, KEGG and Gene Ontology identified up regulation of genes associated with mitotic spindle and G2/M G1/S DNA damage spindle checkpoints, homologous recombination and the cell cycle in the cell lines compared to patient-derived cultures (p<0.01). These data are consistent with the quicker doubling time (DT) and increased sensitivity of ES cell lines to chemotherapy (actinomycin D, doxorubicin, vincristine, etoposide; DT range=19-33h, EC50 range=4-34nM) compared to patient-derived cultures (DT range=55-197h, EC50>10μM, p<0.05). All patient-derived ES cultures formed spheroids of similar diameter (range 105-258µm) at 96h (p>0.05), although there was heterogeneity in spheroid weight (range 205-3639ng, p<0.05) and migration index (MI; range 30-170, p<0.05). Treatment with actinomycin D (1-10 µM) significantly increased spheroid diameter (p<0.05), whereas 4-hyperoxyifosfamide (active metabolite of ifosfamide, 7-30µM) decreased spheroid diameter (p<0.05). Doxorubicin did not significantly change the diameter of spheroids. When propagated in combination with mesenchymal stem cells there was a 2-fold increase in viable cell number (p<0.0001) compared to ES cultures alone. Patient-derived ES cultures in zebrafish were observed at the injection site (perivitelline space). Some cells were detected in circulation 3 days post injection. This is consistent with heterogeneity and the high MI of patient-derived ES cultures. The transcriptome of ES cell lines is significantly different from that of ES patient-derived cultures and is associated with increased doubling times and response to chemotherapies. This highlights the importance of developing preclinical models using patient-derived ES cultures. Patient-derived ES cultures in 2D and 3D models in vitro and in zebrafish may provide a reliable cost-effective preclinical pipeline 1Minas et al. 2017.Oncotarget,8(21):34141-34163.2Nanni et al. 2019.Sci Rep,9(1):12174 3Roundhill et al. 2021.Cell Oncol,44(5):1065-1085. Citation Format: Elizabeth A. Roundhill, Elton J. Vasconcelos, David R. Westhead, Sarah Grissenberger, Martin Distel, Susan A. Burchill. Developing human Ewing sarcoma in vitro models to prioritise new treatments. [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 4683.

Multiple myeloma (MM) shows constitutive activation of canonical and noncanonical nuclear factor & kappa;B (NF-& kappa;B) signaling via genetic mutations or tumor microenvironment (TME) stimulations. A subset of MM cell lines showed dependency for cell growth and survival on the canonical NF-& kappa;B transcription factor RELA alone, suggesting a critical role for a RELA-mediated biological program in MM pathogenesis. Here, we determined the RELA-dependent transcriptional program in MM cell lines and found the expression of the cell surface molecules interleukin-27 receptor-& alpha; (IL-27R & alpha;) and the adhesion molecule JAM2 to be responsive to RELA at the messenger RNA and protein levels. IL-27R & alpha; and JAM2 were expressed on primary MM cells at higher levels than on healthy long-lived plasma cells (PCs) in the bone marrow. IL-27 activated STAT1, and to a lesser extent STAT3, in MM cell lines and in PCs generated from memory B cells in an IL-21-dependent in vitro PC differentiation assay. Concomitant activity of IL-21 and IL-27 enhanced differentiation into PCs and increased the cell-surface expression of the known STAT target gene CD38. In accordance, a subset of MM cell lines and primary MM cells cultured with IL-27 upregulated CD38 cell-surface expression, a finding with potential implications for enhancing the efficacy of CD38-directed monoclonal antibody therapies by increasing CD38 expression on tumor cells. The elevated expression of IL-27R & alpha; and JAM2 on MM cells compared with that on healthy PCs may be exploited for the development of targeted therapeutic strategies that modulate the interaction of MM cells with the TME.

ABSTRACT Human B cell differentiation into antibody secreting plasma cells is a critical process in the adaptive immune response, whose regulation at the genetic level remains incompletely understood. To reveal the temporal sequence of transcription factor driven cellular changes we generated chromatin accessibility (ATAC-seq) and gene expression (RNA-seq) data from in vitro differentiation of human B cells into plasma cells using a published protocol for differentiation up to the plasma cell stage. Using a new computational method, cisREAD (cis-Regulatory Elements Across Differentiation), we defined a core set of cis -regulatory elements that are confidently linked to dynamic transcription factor binding and changes in gene expression across the mature B lineage. Here we describe how cisREAD identifies regulatory element ‘communities’, based on chromatin accessibility and transcription factor co-occupancy, and prioritizes those whose accessibility predicts differential gene expression through regularized regression models. Through downstream analyses of cisREAD-predicted regulation, we show how transcription factors reshape B cell epigenomes and transcriptomes in response to differentiation stimuli. Our results confirm roles for OCT2, IRF4 and PRDM1 in plasma cell differentiation, and reveal that a shift from PU.1/SPIB-driven to AP-1-driven gene regulation is a key determinant of B cell activation. GRAPHICAL ABSTRACT Integration of epigenomic and transcriptomic datasets with the cisREAD method, followed by clustering and network analysis, reveals that gene regulation shifts from PU.1/SPIB to AP-1 upon B cell activation.

Organismal proteostasis is maintained by intercellular signaling processes including cell nonautonomous stress responses such as transcellular chaperone signaling (TCS). When TCS is activated upon tissue-specific knockdown of hsp-90 in the Caenorhabditis elegans intestine, heat-inducible hsp-70 is induced in muscle cells at the permissive temperature resulting in increased heat stress resistance and lifespan extension. However, our understanding of the molecular mechanism and signaling factors mediating transcellular activation of hsp-70 expression from one tissue to another is still in its infancy. Here, we conducted a combinatorial approach using transcriptome RNA-Seq profiling and a forward genetic mutagenesis screen to elucidate how stress signaling from the intestine to the muscle is regulated. We find that the TCS-mediated "gut-to-muscle" induction of hsp-70 expression is suppressed by HSF-1 and instead relies on transcellular-X-cross-tissue (txt) genes. We identify a key role for the PDZ-domain guanylate cyclase txt-1 and the homeobox transcription factor ceh-58 as signaling hubs in the stress receiving muscle cells to initiate hsp-70 expression and facilitate TCS-mediated heat stress resistance and lifespan extension. Our results provide a new view on cell-nonautonomous regulation of "inter-tissue" stress responses in an organism that highlight a key role for the gut. Our data suggest that the HSF-1-mediated heat shock response is switched off upon TCS activation, in favor of an intercellular stress-signaling route to safeguard survival.