Refine
Year of publication
Document Type
- Article (5338)
- Doctoral Thesis (1510)
- Part of Periodical (211)
- Conference Proceeding (189)
- Preprint (171)
- Book (86)
- Contribution to a Periodical (67)
- Review (50)
- Working Paper (22)
- Part of a Book (17)
Language
Keywords
- inflammation (81)
- COVID-19 (60)
- SARS-CoV-2 (48)
- apoptosis (39)
- Inflammation (38)
- cancer (38)
- glioblastoma (38)
- breast cancer (34)
- autophagy (29)
- prostate cancer (29)
Institute
- Medizin (7681) (remove)
The use of artificial intelligence (AI) systems in biomedical and clinical settings can disrupt the traditional doctor–patient relationship, which is based on trust and transparency in medical advice and therapeutic decisions. When the diagnosis or selection of a therapy is no longer made solely by the physician, but to a significant extent by a machine using algorithms, decisions become nontransparent. Skill learning is the most common application of machine learning algorithms in clinical decision making. These are a class of very general algorithms (artificial neural networks, classifiers, etc.), which are tuned based on examples to optimize the classification of new, unseen cases. It is pointless to ask for an explanation for a decision. A detailed understanding of the mathematical details of an AI algorithm may be possible for experts in statistics or computer science. However, when it comes to the fate of human beings, this “developer’s explanation” is not sufficient. The concept of explainable AI (XAI) as a solution to this problem is attracting increasing scientific and regulatory interest. This review focuses on the requirement that XAIs must be able to explain in detail the decisions made by the AI to the experts in the field.
Sex differences in pain perception have been extensively studied, but precision medicine applications such as sex-specific pain pharmacology have barely progressed beyond proof-of-concept. A data set of pain thresholds to mechanical (blunt and punctate pressure) and thermal (heat and cold) stimuli applied to non-sensitized and sensitized (capsaicin, menthol) forearm skin of 69 male and 56 female healthy volunteers was analyzed for data structures contingent with the prior sex structure using unsupervised and supervised approaches. A working hypothesis that the relevance of sex differences could be approached via reversibility of the association, i.e., sex should be identifiable from pain thresholds, was verified with trained machine learning algorithms that could infer a person's sex in a 20% validation sample not seen to the algorithms during training, with balanced accuracy of up to 79%. This was only possible with thresholds for mechanical stimuli, but not for thermal stimuli or sensitization responses, which were not sufficient to train an algorithm that could assign sex better than by guessing or when trained with nonsense (permuted) information. This enabled the translation to the molecular level of nociceptive targets that convert mechanical but not thermal information into signals interpreted as pain, which could eventually be used for pharmacological precision medicine approaches to pain. By exploiting a key feature of machine learning, which allows for the recognition of data structures and the reduction of information to the minimum relevant, experimental human pain data could be characterized in a way that incorporates "non" logic that could be translated directly to the molecular pharmacological level, pointing toward sex-specific precision medicine for pain.
Bacteria that are capable of organizing themselves as biofilms are an important public health issue. Knowledge discovery focusing on the ability to swarm and conquer the surroundings to form persistent colonies is therefore very important for microbiological research communities that focus on a clinical perspective. Here, we demonstrate how a machine learning workflow can be used to create useful models that are capable of discriminating distinct associated growth behaviors along distinct phenotypes. Based on basic gray-scale images, we provide a processing pipeline for binary image generation, making the workflow accessible for imaging data from a wide range of devices and conditions. The workflow includes a locally estimated regression model that easily applies to growth-related data and a shape analysis using identified principal components. Finally, we apply a density-based clustering application with noise (DBSCAN) to extract and analyze characteristic, general features explained by colony shapes and areas to discriminate distinct Bacillus subtilis phenotypes. Our results suggest that the differences regarding their ability to swarm and subsequently conquer the medium that surrounds them result in characteristic features. The differences along the time scales of the distinct latency for the colony formation give insights into the ability to invade the surroundings and therefore could serve as a useful monitoring tool.
Knowledge discovery in biomedical data using supervised methods assumes that the data contain structure relevant to the class structure if a classifier can be trained to assign a case to the correct class better than by guessing. In this setting, acceptance or rejection of a scientific hypothesis may depend critically on the ability to classify cases better than randomly, without high classification performance being the primary goal. Random forests are often chosen for knowledge-discovery tasks because they are considered a powerful classifier that does not require sophisticated data transformation or hyperparameter tuning and can be regarded as a reference classifier for tabular numerical data. Here, we report a case where the failure of random forests using the default hyperparameter settings in the standard implementations of R and Python would have led to the rejection of the hypothesis that the data contained structure relevant to the class structure. After tuning the hyperparameters, classification performance increased from 56% to 65% balanced accuracy in R, and from 55% to 67% balanced accuracy in Python. More importantly, the 95% confidence intervals in the tuned versions were to the right of the value of 50% that characterizes guessing-level classification. Thus, tuning provided the desired evidence that the data structure supported the class structure of the data set. In this case, the tuning made more than a quantitative difference in the form of slightly better classification accuracy, but significantly changed the interpretation of the data set. This is especially true when classification performance is low and a small improvement increases the balanced accuracy to over 50% when guessing.
Bayesian inference is ubiquitous in science and widely used in biomedical research such as cell sorting or “omics” approaches, as well as in machine learning (ML), artificial neural networks, and “big data” applications. However, the calculation is not robust in regions of low evidence. In cases where one group has a lower mean but a higher variance than another group, new cases with larger values are implausibly assigned to the group with typically smaller values. An approach for a robust extension of Bayesian inference is proposed that proceeds in two main steps starting from the Bayesian posterior probabilities. First, cases with low evidence are labeled as “uncertain” class membership. The boundary for low probabilities of class assignment (threshold 𝜀
) is calculated using a computed ABC analysis as a data-based technique for item categorization. This leaves a number of cases with uncertain classification (p < 𝜀
). Second, cases with uncertain class membership are relabeled based on the distance to neighboring classified cases based on Voronoi cells. The approach is demonstrated on biomedical data typically analyzed with Bayesian statistics, such as flow cytometric data sets or biomarkers used in medical diagnostics, where it increased the class assignment accuracy by 1–10% depending on the data set. The proposed extension of the Bayesian inference of class membership can be used to obtain robust and plausible class assignments even for data at the extremes of the distribution and/or for which evidence is weak.
In dengue-endemic countries such as Indonesia, Zika may be misdiagnosed as dengue, leading to underestimates of Zika disease and less foreknowledge of pregnancy-related complications such as microcephaly. Objective: To assess the attitudes of frontline physicians in a dengue-endemic country toward testing for Zika infection among patients with dengue-like illnesses. Methods: A cross-sectional online survey was conducted among general practitioners (GPs) in Indonesia. The survey assessed their attitude and also collected sociodemographic data, characteristics of their medical education, professional background, and workplace, and exposure to Zika cases. A two-step logistic regression analysis was used to assess possible variables associated with these attitudes. Results: A total of 370 GPs were included in the final analysis of which 70.8% had good attitude. Unadjusted analyses suggested that GPs who were 30 years old or older and those who had medical experience five years or longer had lower odds of having a positive attitude compared to those who aged younger than 30 years and those who had medical experience less than five years, OR: 0.58; 95%CI: 0.37, 0.91 and OR: 0.55; 95%CI: 0.35, 0.86, respectively. No explanatory variable was associated with attitude in the fully adjusted model. Conclusion: Our findings point to younger GPs with a shorter medical experience being more likely to consider testing for Zika infection among their patients presenting with dengue-like illnesses. Strategic initiatives may be needed to enhance older or longer-experienced physicians' capacity in diagnosing Zika infection.
BH3 mimetics are promising novel anticancer therapeutics. By selectively inhibiting BCL-2, BCL-xL, or MCL-1 (i.e. ABT-199, A-1331852, S63845) they shift the balance of pro- and anti-apoptotic proteins in favor of apoptosis. As Bromodomain and Extra Terminal (BET) protein inhibitors promote pro-apoptotic rebalancing, we evaluated the potential of the BET inhibitor JQ1 in combination with ABT-199, A-1331852 or S63845 in rhabdomyosarcoma (RMS) cells. The strongest synergistic interaction was identified for JQ1/A-1331852 and JQ1/S63845 co-treatment, which reduced cell viability and long-term clonogenic survival. Mechanistic studies revealed that JQ1 upregulated BIM and NOXA accompanied by downregulation of BCL-xL, promoting pro-apoptotic rebalancing of BCL-2 proteins. JQ1/A-1331852 and JQ1/S63845 co-treatment enhanced this pro-apoptotic rebalancing and triggered BAK- and BAX-dependent apoptosis since a) genetic silencing of BIM, BAK or BAX, b) inhibition of caspase activity with zVAD.fmk and c) overexpression of BCL-2 all rescued JQ1/A-1331852- and JQ1/S63845-induced cell death. Interestingly, NOXA played a different role in both treatments, as genetic silencing of NOXA significantly rescued from JQ1/A-1331852-mediated apoptosis but not from JQ1/S63845-mediated apoptosis. In summary, JQ1/A-1331852 and JQ1/S63845 co-treatment represent new promising therapeutic strategies to synergistically trigger mitochondrial apoptosis in RMS.
Burkitt's lymphoma (BL) is a highly aggressive form of B-cell non-Hodgkin's lymphoma. The clinical outcome in children with BL has improved over the last years but the prognosis for adults is still poor, highlighting the need for novel treatment strategies. Here, we report that the combinational treatment with the Smac mimetic BV6 and TRAIL triggers necroptosis in BL when caspases are blocked by zVAD.fmk (TBZ treatment). The sensitivity of BL cells to TBZ correlates with MLKL expression. We demonstrate that necroptotic signaling critically depends on MLKL, since siRNA-induced knockdown and CRISPR/Cas9-mediated knockout of MLKL profoundly protect BL cells from TBZ-induced necroptosis. Conversely, MLKL overexpression in cell lines expressing low levels of MLKL leads to necroptosis induction, which can be rescued by pharmacological inhibitors, highlighting the important role of MLKL for necroptosis execution. Importantly, the methylation status analysis of the MLKL promoter reveals a correlation between methylation and MLKL expression. Thus, MLKL is epigenetically regulated in BL and might serve as a prognostic marker for treatment success of necroptosis-based therapies. These findings have crucial implications for the development of new treatment options for BL.
The evaluation of pharmacological data using machine learning requires high data quality. Therefore, data preprocessing, that is, cleaning analytical laboratory errors, replacing missing values or outliers, and transforming data adequately before actual data analysis, is crucial. Because current tools available for this purpose often require programming skills, preprocessing tools with graphical user interfaces that can be used interactively are needed. In collaboration between data scientists and experts in bioanalytical diagnostics, a graphical software package for data preprocessing called pguIMP is proposed, which contains a fixed sequence of preprocessing steps to enable reproducible interactive data preprocessing. As an R-based package, it also allows direct integration into this data science environment without requiring any programming knowledge. The implementation of contemporary data processing methods, including machine-learning-based imputation techniques, ensures the generation of corrected and cleaned bioanalytical data sets that preserve data structures such as clusters better than is possible with classical methods. This was evaluated on bioanalytical data sets from lipidomics and drug research using k-nearest-neighbors-based imputation followed by k-means clustering and density-based spatial clustering of applications with noise. The R package provides a Shiny-based web interface designed to be easy to use for non–data analysis experts. It is demonstrated that the spectrum of methods provided is suitable as a standard pipeline for preprocessing bioanalytical data in biomedical research domains. The R package pguIMP is freely available at the comprehensive R archive network (https://cran.r-project.org/web/packages/pguIMP/index.html).
Background: Eukaryotic gene expression is controlled by cis-regulatory elements (CREs), including promoters and enhancers, which are bound by transcription factors (TFs). Differential expression of TFs and their binding affinity at putative CREs determine tissue- and developmental-specific transcriptional activity. Consolidating genomic data sets can offer further insights into the accessibility of CREs, TF activity, and, thus, gene regulation. However, the integration and analysis of multi-modal data sets are hampered by considerable technical challenges. While methods for highlighting differential TF activity from combined chromatin state data (e.g., ChIP-seq, ATAC-seq, or DNase-seq) and RNA-seq data exist, they do not offer convenient usability, have limited support for large-scale data processing, and provide only minimal functionality for visually interpreting results.
Results: We developed TF-Prioritizer, an automated pipeline that prioritizes condition-specific TFs from multi-modal data and generates an interactive web report. We demonstrated its potential by identifying known TFs along with their target genes, as well as previously unreported TFs active in lactating mouse mammary glands. Additionally, we studied a variety of ENCODE data sets for cell lines K562 and MCF-7, including twelve histone modification ChIP-seq as well as ATAC-seq and DNase-seq datasets, where we observe and discuss assay-specific differences.
Conclusion: TF-Prioritizer accepts ATAC-seq, DNase-seq, or ChIP-seq and RNA-seq data as input and identifies TFs with differential activity, thus offering an understanding of genome-wide gene regulation, potential pathogenesis, and therapeutic targets in biomedical research.
Background Eukaryotic gene expression is controlled by cis-regulatory elements (CREs) including promoters and enhancers which are bound by transcription factors (TFs). Differential expression of TFs and their putative binding sites on CREs cause tissue and developmental-specific transcriptional activity. Consolidating genomic data sets can offer further insights into the accessibility of CREs, TF activity, and thus gene regulation. However, the integration and analysis of multi-modal data sets are hampered by considerable technical challenges. While methods for highlighting differential TF activity from combined ChIP-seq and RNA-seq data exist, they do not offer good usability, have limited support for large-scale data processing, and provide only minimal functionality for visual result interpretation.
Results We developed TF-Prioritizer, an automated java pipeline to prioritize condition-specific TFs derived from multi-modal data. TF-Prioritizer creates an interactive, feature-rich, and user-friendly web report of its results. To showcase the potential of TF-Prioritizer, we identified known active TFs (e.g., Stat5, Elf5, Nfib, Esr1), their target genes (e.g., milk proteins and cell-cycle genes), and newly classified lactating mammary gland TFs (e.g., Creb1, Arnt).
Conclusion TF-Prioritizer accepts ChIP-seq and RNA-seq data, as input and suggests TFs with differential activity, thus offering an understanding of genome-wide gene regulation, potential pathogenesis, and therapeutic targets in biomedical research.
Co-targeting MCL-1 and ERK1/2 kinase induces mitochondrial apoptosis in rhabdomyosarcoma cells
(2021)
The RAS/MEK/ERK genetic axis is commonly altered in rhabdomyosarcoma (RMS), indicating high activity of downstream effector ERK1/2 kinase. Previously, we have demonstrated that inhibition of the RAS/MEK/ERK signaling pathway in RMS is insufficient to induce cell death due to residual pro-survival MCL-1 activity. Here, we show that the combination of ERK1/2 inhibitor Ulixertinib and MCL-1 inhibitor S63845 is highly synergistic and induces apoptotic cell death in RMS in vitro and in vivo. Importantly, Ulixertinib/S63845 co-treatment suppresses long-term survival of RMS cells, induces rapid caspase activation and caspase-dependent apoptosis. Mechanistically, Ulixertinib-mediated upregulation of BIM and BMF in combination with MCL-1 inhibition by S63845 shifts the balance of BCL-2 proteins towards a pro-apoptotic state resulting in apoptosis induction. A genetic silencing approach reveals that BIM, BMF, BAK and BAX are all required for Ulixertinib/S63845-induced apoptosis. Overexpression of BCL-2 rescues cell death triggered by Ulixertinib/S63845 co-treatment, confirming that combined inhibition of ERK1/2 and MCL-1 effectively induces cell death of RMS cells via the intrinsic mitochondrial apoptotic pathway. Thus, this study is the first to demonstrate the cytotoxic potency of co-inhibition of ERK1/2 and MCL-1 for RMS treatment.
This survey reports on the DNA identification and occurrence of Culex torrentium and Cx. pipiens s.s. in Belgium. These native disease-vector mosquito species are morphologically difficult to separate, and the biotypes of Cx. pipiens s.s. are morphologically indistinguishable. Culex torrentium and Cx. pipiens s.s. were identified using the COI and ACE2 loci. We recorded 1248 Cx. pipiens s.s. and 401 Cx. torrentium specimens from 24 locations in Belgium (collected between 2017 and 2019). Culex pipiens biotypes pipiens and molestus, and their hybrids, were differentiated using fragment-size analysis of the CQ11 locus (956 pipiens and 227 molestus biotype specimens, 29 hybrids). Hybrids were observed at 13 out of 16 sympatric sites. These results confirm that both species are widespread in Belgium, but while Cx. torrentium revealed many COI haplotypes, Cx. pipiens s.s. showed only one abundant haplotype. This latter observation may either reflect a recent population-wide demographic or range expansion, or a recent bottleneck, possibly linked to a Wolbachia infection. Finally, new evidence is provided for the asymmetric but limited introgression of the molestus biotype into the pipiens biotype.
Most sRNA biogenesis mechanisms involve either RNAseIII cleavage or ping-pong amplification by different Piwi proteins harboring slicer activity. Here, we follow the question why the mechanism of transgene-induced silencing in the ciliate Paramecium needs both Dicer activity and two Ptiwi proteins. This pathway involves primary siRNAs produced from non-translatable transgenes and secondary siRNAs from endogenous remote loci. Our data does not indicate any signatures from ping-pong amplification but Dicer cleavage of long dsRNA. We show that Ptiwi13 and 14 have different preferences for primary and secondary siRNAs but do not load them mutually exclusive. Both Piwis enrich for antisense RNAs and Ptiwi14 loaded siRNAs show a 5′-U signature. Both Ptiwis show in addition a general preference for Uridine-rich sRNAs along the entire sRNA length. Our data indicates both Ptiwis and 2’-O-methylation to contribute to strand selection of Dicer cleaved siRNAs. This unexpected function of two distinct vegetative Piwis extends the increasing knowledge of the diversity of Piwi functions in diverse silencing pathways. As both Ptiwis show differential subcellular localisation, Ptiwi13 in the cytoplasm and Ptiwi14 in the vegetative macronucleus, we conclude that cytosolic and nuclear silencing factors are necessary for efficient chromatin silencing.
The unicellular ciliate Paramecium contains a large vegetative macronucleus with several unusual characteristics, including an extremely high coding density and high polyploidy. As macronculear chromatin is devoid of heterochromatin, our study characterizes the functional epigenomic organization necessary for gene regulation and proper Pol II activity. Histone marks (H3K4me3, H3K9ac, H3K27me3) reveal no narrow peaks but broad domains along gene bodies, whereas intergenic regions are devoid of nucleosomes. Our data implicate H3K4me3 levels inside ORFs to be the main factor associated with gene expression, and H3K27me3 appears in association with H3K4me3 in plastic genes. Silent and lowly expressed genes show low nucleosome occupancy, suggesting that gene inactivation does not involve increased nucleosome occupancy and chromatin condensation. Because of a high occupancy of Pol II along highly expressed ORFs, transcriptional elongation appears to be quite different from that of other species. This is supported by missing heptameric repeats in the C-terminal domain of Pol II and a divergent elongation system. Our data imply that unoccupied DNA is the default state, whereas gene activation requires nucleosome recruitment together with broad domains of H3K4me3. In summary, gene activation and silencing in Paramecium run counter to the current understanding of chromatin biology.
Background: Blood donation saves lives. Provided they are in good health, male volunteers can donate as often as six times per year from the age of 18 into their late sixties. The burden of blood donation is very unevenly distributed, with a small minority of altruistic individuals providing this critical resource. While the consequences of persistent iron depletion in blood donors have been studied in the context of cancer and coronary heart disease, potential effects of the erythropoietic stress from repetitive large-volume phlebotomy remain unexplored. We sought to investigate if and how repeated blood donations affect the clonal composition of the hematopoietic stem and progenitor cell (HSPC) compartment.
Methods: 105 healthy, male individuals with an extensive blood donation history (median of 120 donations per donor; median age of 66 yrs.) were screened for the presence of clonal hematopoiesis (CH) using a sequencing panel covering 141 genes commonly mutated in human myeloid neoplasms. The control cohort consisted of 103 healthy, male donors with a median of 5 donations per donor and a median age of 63. Donors positive for CH were subsequently studied longitudinally. The pathogenicity of detected variants was compared using established scoring systems. Finally, to assess the functional consequences of blood-donation induced CH, selected CH mutations were introduced by CRISPR-mediated editing into HSPCs from human cord blood (CB) or bone marrow (BM). The effect of these mutations was tested under different stress stimuli using functional ex vivo long-term culture initiating cells (LTC-IC) assays.
Results: Compared to the control cohort, frequent donors were significantly more likely to have mutations in genes encoding for epigenetic modifiers (44.7 vs. 22.3 %), most specifically in the two genes most commonly mutated in CH, DNMT3A and TET2 (35.2 vs. 20.3 %). However, no difference in the variant allele frequency (VAF) of detected mutations was found between the groups. Longitudinal analysis revealed that the majority of the mutations remained at a stable VAF over an observation period of approximately one year. Three DNMT3A variants from the frequent donor cohort were introduced into healthy HSPCs and functionally analyzed: All expanded in response to EPO, but none responded to LPS or IFNγ stimulation. This contrasted with the leukemogenic DNMT3A R882H mutation, which did not expand in the presence of EPO but instead responded strongly to inflammatory stimuli.
Conclusions: Frequent whole blood donation is associated with a higher prevalence of CH driven by mutations in genes encoding for epigenetic modifiers, with DNMT3A and TET2 being the most common. This increased CH prevalence is not associated with a higher pathogenicity of the associated variants and is likely a result of the selection of clones with improved responsiveness to EPO under the condition of bleeding stress. Our data show that even highly frequent blood donations over many years is not increasing the risk for malignant clones further underscoring the safety of repetitive blood donations. To our knowledge, this is the first CH study analyzing a cohort of individuals known for their superior health and survival, able to donate blood until advanced age. Thus, our analysis possibly identified mutations associated with beneficial outcomes, rather than a disease condition, such as mutations in DNMT3A that mediated the improved expansion of HSPCs in EPO enriched environments. Our data support the notion of ongoing Darwinian evolution in humans at the somatic stem cell level and present EPO as one of the environmental factors to which HSPCs with specific mutations may respond with superior fitness.
Objective: This paper presents a novel digital workflow that expedites and facilitates the manufacturing of high-end full-ceramic restorations based on “Print and Press”-Technology combined with 3D-printed colored 3D-models.
Clinical considerations: Despite ongoing innovations and developments in the digital workflow, the precision, and the final esthetic outcome is still limited compared with conventional press ceramics. The proposed method combines the advantages of digital scan- and design technologies with the proven conventional press-technology to accomplish high-end full-ceramic restorations. The restoration is digitally designed, the data set is 3D-printed in resin that can be burned out, subsequently conventionally embedded and pressed. Final esthetic finishing of the partial restorations is done on a 3D-printed physical colored 3D-model.
Conclusion: The report describes synergetic effects of digital and analog procedures. 3D-printed colored 3D-models can positively support the manufacturing of full ceramic restorations regarding their optical integration. Therefore, the use of 3D-printed colored 3D-models signifies a new innovative technique with many promising application areas.
Clinical significance: The combination of excellent clinical long-term data for pressed ceramic restorations and proven digital processes, like intraoral scanning, design, and additive manufacturing, in the dental field promise an individual workflow for predictability and excellent esthetics.
Hidradenitis suppurativa (HS) is a chronic inflammatory skin disease of the hair follicles leading to painful lesions, associated with increased levels of pro-inflammatory cytokines. Numerous guidelines recommend antibiotics like clindamycin and rifampicin in combination, as first-line systemic therapy in moderate-to-severe forms of inflammation. HS has been proposed to be mainly an auto-inflammatory disease associated with but not initially provoked by bacteria. Therefore, it has to be assumed that the pro-inflammatory milieu previously observed in HS skin is not solely dampened by the bacteriostatic inhibition of DNA-dependent RNA polymerase. To further clarify the mechanism of anti-inflammatory effects of rifampicin, ex vivo explants of lesional HS from 8 HS patients were treated with rifampicin, and its effect on cytokine production, immune cells as well as the expression of Toll-like receptor 2 (TLR2) were investigated. Analysis of cell culture medium of rifampicin-treated HS explants revealed an anti-inflammatory effect of rifampicin that significantly inhibiting interleukin (IL)-1β, IL-6, IL-8, IL-10 and tumour necrosis factor (TNF)-α production. Immunohistochemistry of the rifampicin-treated explants suggested a tendency for it to reduce the expression of TLR2 while not affecting the number of immune cells.
In gastric cancer (GC), there are four molecular subclasses that indicate whether patients respond to chemotherapy or immunotherapy, according to the TCGA. In clinical practice, however, not every patient undergoes molecular testing. Many laboratories have used well-implemented in situ techniques (IHC and EBER-ISH) to determine the subclasses in their cohorts. Although multiple stains are used, we show that a staining approach is unable to correctly discriminate all subclasses. As an alternative, we trained an ensemble convolutional neuronal network using bagging that can predict the molecular subclass directly from hematoxylin–eosin histology. We also identified patients with predicted intra-tumoral heterogeneity or with features from multiple subclasses, which challenges the postulated TCGA-based decision tree for GC subtyping. In the future, deep learning may enable targeted testing for molecular subtypes and targeted therapy for a broader group of GC patients. © 2022 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.