Adaptive digital tissue deconvolution.

MOTIVATION: The inference of cellular compositions from bulk and spatial transcriptomics data increasingly complements data analyses. Multiple computational approaches were suggested and recently, machine learning techniques were developed to systematically improve estimates. Such approaches allow to infer additional, less abundant cell types. However, they rely on training data which do not capture the full biological diversity encountered in transcriptomics analyses; data can contain cellular contributions not seen in the training data and as such, analyses can be biased or blurred. Thus, computational approaches have to deal with unknown, hidden contributions. Moreover, most methods are based on cellular archetypes which serve as a reference; e.g. a generic T-cell profile is used to infer the proportion of T-cells. It is well known that cells adapt their molecular phenotype to the environment and that pre-specified cell archetypes can distort the inference of cellular compositions. RESULTS: We propose Adaptive Digital Tissue Deconvolution (ADTD) to estimate cellular proportions of pre-selected cell types together with possibly unknown and hidden background contributions. Moreover, ADTD adapts prototypic reference profiles to the molecular environment of the cells, which further resolves cell-type specific gene regulation from bulk transcriptomics data. We verify this in simulation studies and demonstrate that ADTD improves existing approaches in estimating cellular compositions. In an application to bulk transcriptomics data from breast cancer patients, we demonstrate that ADTD provides insights into cell-type specific molecular differences between breast cancer subtypes. AVAILABILITY AND IMPLEMENTATION: A python implementation of ADTD and a tutorial are available at Gitlab and zenodo (doi:10.5281/zenodo.7548362).

CODEX: COunterfactual Deep learning for the in silico EXploration of cancer cell line perturbations.

MOTIVATION: High-throughput screens (HTS) provide a powerful tool to decipher the causal effects of chemical and genetic perturbations on cancer cell lines. Their ability to evaluate a wide spectrum of interventions, from single drugs to intricate drug combinations and CRISPR-interference, has established them as an invaluable resource for the development of novel therapeutic approaches. Nevertheless, the combinatorial complexity of potential interventions makes a comprehensive exploration intractable. Hence, prioritizing interventions for further experimental investigation becomes of utmost importance. RESULTS: We propose CODEX (COunterfactual Deep learning for the in silico EXploration of cancer cell line perturbations) as a general framework for the causal modeling of HTS data, linking perturbations to their downstream consequences. CODEX relies on a stringent causal modeling strategy based on counterfactual reasoning. As such, CODEX predicts drug-specific cellular responses, comprising cell survival and molecular alterations, and facilitates the in silico exploration of drug combinations. This is achieved for both bulk and single-cell HTS. We further show that CODEX provides a rationale to explore complex genetic modifications from CRISPR-interference in silico in single cells. AVAILABILITY AND IMPLEMENTATION: Our implementation of CODEX is publicly available at https://github.com/sschrod/CODEX. All data used in this article are publicly available.

Stable feature selection utilizing Graph Convolutional Neural Network and Layer-wise Relevance Propagation for biomarker discovery in breast cancer.

High-throughput technologies are becoming increasingly important in discovering prognostic biomarkers and in identifying novel drug targets. With Mammaprint, Oncotype DX, and many other prognostic molecular signatures breast cancer is one of the paradigmatic examples of the utility of high-throughput data to deliver prognostic biomarkers, that can be represented in a form of a rather short gene list. Such gene lists can be obtained as a set of features (genes) that are important for the decisions of a Machine Learning (ML) method applied to high-dimensional gene expression data. Several studies have identified predictive gene lists for patient prognosis in breast cancer, but these lists are unstable and have only a few genes in common. Instability of feature selection impedes biological interpretability: genes that are relevant for cancer pathology should be members of any predictive gene list obtained for the same clinical type of patients. Stability and interpretability of selected features can be improved by including information on molecular networks in ML methods. Graph Convolutional Neural Network (GCNN) is a contemporary deep learning approach applicable to gene expression data structured by a prior knowledge molecular network. Layer-wise Relevance Propagation (LRP) and SHapley Additive exPlanations (SHAP) are methods to explain individual decisions of deep learning models. We used both GCNN+LRP and GCNN+SHAP techniques to construct feature sets by aggregating individual explanations. We suggest a methodology to systematically and quantitatively analyze the stability, the impact on the classification performance, and the interpretability of the selected feature sets. We used this methodology to compare GCNN+LRP to GCNN+SHAP and to more classical ML-based feature selection approaches. Utilizing a large breast cancer gene expression dataset we show that, while feature selection with SHAP is useful in applications where selected features have to be impactful for classification performance, among all studied methods GCNN+LRP delivers the most stable (reproducible) and interpretable gene lists.

Histological assessment of cardiac amyloidosis in patients undergoing transcatheter aortic valve replacement.

AIMS: Studies have reported a strongly varying co-prevalence of aortic stenosis (AS) and cardiac amyloidosis (CA). We sought to histologically determine the co-prevalence of AS and CA in patients undergoing transcatheter aortic valve replacement (TAVR). Consequently, we aimed to derive an algorithm to identify cases in which to suspect the co-prevalence of AS and CA. METHODS AND RESULTS: In this prospective, monocentric study, endomyocardial biopsies of 162 patients undergoing TAVR between January 2017 and March 2021 at the University Medical Centre Göttingen were analysed by one pathologist blinded to clinical data using haematoxylin-eosin staining, Elastica van Gieson staining, and Congo red staining of endomyocardial biopsies. CA was identified in only eight patients (4.9%). CA patients had significantly higher N-terminal pro-brain natriuretic peptide (NT-proBNP) levels (4356.20 vs. 1938.00 ng/L, P = 0.034), a lower voltage-to-mass ratio (0.73 vs. 1.46 × 10-2 mVm2/g, P = 0.022), and lower transaortic gradients (Pmean 17.5 vs. 38.0 mmHg, P = 0.004) than AS patients. Concomitant CA was associated with a higher prevalence of post-procedural acute kidney injury (50.0% vs. 13.1%, P = 0.018) and sudden cardiac death [SCD; P (log-rank test) = 0.017]. Following propensity score matching, 184 proteins were analysed to identify serum biomarkers of concomitant CA. CA patients expressed lower levels of chymotrypsin (P = 0.018) and carboxypeptidase 1 (P = 0.027). We propose an algorithm using commonly documented parameters-stroke volume index, ejection fraction, NT-proBNP levels, posterior wall thickness, and QRS voltage-to-mass ratio-to screen for CA in AS patients, reaching a sensitivity of 66.6% with a specificity of 98.1%. CONCLUSIONS: The co-prevalence of AS and CA was lower than expected, at 4.9%. Despite excellent 1 year mortality, AS + CA patients died significantly more often from SCD. We propose a multimodal algorithm to facilitate more effective screening for CA containing parameters commonly documented during clinical routine. Proteomic biomarkers may yield additional information in the future.

Impact of Portal Vein Resection (PVR) in Patients Who Underwent Curative Intended Pancreatic Head Resection.

The oncological impact of portal vein resection (PVR) in pancreatic cancer surgery remains contradictory. Different variables might have an impact on the outcome. The aim of the present study is the retrospective assessment of the frequency of PVR, histological confirmation of tumor infiltration, and comparison of oncological outcomes in PVR patients. We retrieved n = 90 patients from a prospectively collected data bank who underwent pancreas surgery between 2012 and 2019 at the University Medical Centre Göttingen (Germany) and showed a histologically confirmed pancreatic ductal adenocarcinoma (PDAC). While 50 patients (55.6%) underwent pancreatic resection combined with PVR, 40 patients (44.4%) received standard pancreatic surgery. Patients with distal pancreatectomy or a tumor other than PDAC were excluded. PVR was performed either as local excision or circular resection of the portal vein. Clinical/patient data and follow-ups were retrieved. The median follow-up period was 20.5 months. Regarding the oncological outcome, a statistically poorer CSS (p = 0.04) was observed in PVR patients. There was no difference (p = 0.18) in patients' outcomes between tangential and complete PVR, while n = 21 (42% of PVR patients) showed portal vein infiltration. The correlation between performed PVR and resection status was statistically significant: 48.6% of PVR patients achieved R0 resections compared to 75% in non-PVR patients (p = 0.03). Patients who underwent PDAC surgery with PVR show a significantly poorer outcome regardless of PVR type. Tumor size and R-status remain two important variables significantly associated with outcome. Since there is a lack of standardization for the indication of PVR, it remains unknown if the need for resection of vein structures during pancreatic resection represents the biological aggressiveness of the tumor or is biased by the experience of the surgeon.

Ensemble-GNN: federated ensemble learning with graph neural networks for disease module discovery and classification.

SUMMARY: Federated learning enables collaboration in medicine, where data is scattered across multiple centers without the need to aggregate the data in a central cloud. While, in general, machine learning models can be applied to a wide range of data types, graph neural networks (GNNs) are particularly developed for graphs, which are very common in the biomedical domain. For instance, a patient can be represented by a protein-protein interaction (PPI) network where the nodes contain the patient-specific omics features. Here, we present our Ensemble-GNN software package, which can be used to deploy federated, ensemble-based GNNs in Python. Ensemble-GNN allows to quickly build predictive models utilizing PPI networks consisting of various node features such as gene expression and/or DNA methylation. We exemplary show the results from a public dataset of 981 patients and 8469 genes from the Cancer Genome Atlas (TCGA). AVAILABILITY AND IMPLEMENTATION: The source code is available at https://github.com/pievos101/Ensemble-GNN, and the data at Zenodo (DOI: 10.5281/zenodo.8305122).

Onkopipe: A Snakemake Based DNA-Sequencing Pipeline for Clinical Variant Analysis in Precision Medicine.

NGS is increasingly used in precision medicine, but an automated sequencing pipeline that can detect different types of variants (single nucleotide - SNV, copy number - CNV, structural - SV) and does not rely on normal samples as germline comparison is needed. To address this, we developed Onkopipe, a Snakemake-based pipeline that integrates quality control, read alignments, BAM pre-processing, and variant calling tools to detect SNV, CNV, and SV in a unified VCF format without matched normal samples. Onkopipe is containerized and provides features such as reproducibility, parallelization, and easy customization, enabling the analysis of genomic data in precision medicine. Our validation and evaluation demonstrate high accuracy and concordance, making Onkopipe a valuable open-source resource for molecular tumor boards. Onkopipe is being shared as an open source project and is available at https://gitlab.gwdg.de/MedBioinf/mtb/onkopipe.

An overview and a roadmap for artificial intelligence in hematology and oncology.

BACKGROUND: Artificial intelligence (AI) is influencing our society on many levels and has broad implications for the future practice of hematology and oncology. However, for many medical professionals and researchers, it often remains unclear what AI can and cannot do, and what are promising areas for a sensible application of AI in hematology and oncology. Finally, the limits and perils of using AI in oncology are not obvious to many healthcare professionals. METHODS: In this article, we provide an expert-based consensus statement by the joint Working Group on "Artificial Intelligence in Hematology and Oncology" by the German Society of Hematology and Oncology (DGHO), the German Association for Medical Informatics, Biometry and Epidemiology (GMDS), and the Special Interest Group Digital Health of the German Informatics Society (GI). We provide a conceptual framework for AI in hematology and oncology. RESULTS: First, we propose a technological definition, which we deliberately set in a narrow frame to mainly include the technical developments of the last ten years. Second, we present a taxonomy of clinically relevant AI systems, structured according to the type of clinical data they are used to analyze. Third, we show an overview of potential applications, including clinical, research, and educational environments with a focus on hematology and oncology. CONCLUSION: Thus, this article provides a point of reference for hematologists and oncologists, and at the same time sets forth a framework for the further development and clinical deployment of AI in hematology and oncology in the future.

Molecular differences of angiogenic versus vessel co-opting colorectal cancer liver metastases at single-cell resolution.

BACKGROUND: Colorectal cancer liver metastases (CRCLM) are associated with a poor prognosis, reflected by a five-year survival rate of 14%. Anti-angiogenic therapy through anti-VEGF antibody administration is one of the limited therapies available. However, only a subgroup of metastases uses sprouting angiogenesis to secure their nutrients and oxygen supply, while others rely on vessel co-option (VCO). The distinct mode of vascularization is reflected by specific histopathological growth patterns (HGPs), which have proven prognostic and predictive significance. Nevertheless, their molecular mechanisms are poorly understood. METHODS: We evaluated CRCLM from 225 patients regarding their HGP and clinical data. Moreover, we performed spatial (21,804 spots) and single-cell (22,419 cells) RNA sequencing analyses to explore molecular differences in detail, further validated in vitro through immunohistochemical analysis and patient-derived organoid cultures. RESULTS: We detected specific metabolic alterations and a signature of WNT signalling activation in metastatic cancer cells related to the VCO phenotype. Importantly, in the corresponding healthy liver of CRCLM displaying sprouting angiogenesis, we identified a predominantly expressed capillary subtype of endothelial cells, which could be further explored as a possible predictor for HGP relying on sprouting angiogenesis. CONCLUSION: These findings may prove to be novel therapeutic targets to the treatment of CRCLM, in special the ones relying on VCO.

Intralesional TLR4 agonist treatment strengthens the organ defense against colonizing cancer cells in the brain.

Brain metastasis in breast cancer remains difficult to treat and its incidence is increasing. Therefore, the development of new therapies is of utmost clinical relevance. Recently, toll-like receptor (TLR) 4 was correlated with IL6 expression and poor prognosis in 1 215 breast cancer primaries. In contrast, we demonstrated that TLR4 stimulation reduces microglia-assisted breast cancer cell invasion. However, the expression, prognostic value, or therapeutic potential of TLR signaling in breast cancer brain metastasis have not been investigated. We thus tested the prognostic value of various TLRs in two brain-metastasis gene sets. Furthermore, we investigated different TLR agonists, as well as MyD88 and TRIF-deficient microenvironments in organotypic brain-slice ex vivo co-cultures and in vivo colonization experiments. These experiments underline the ambiguous roles of TLR4, its adapter MyD88, and the target nitric oxide (NO) during brain colonization. Moreover, analysis of the gene expression datasets of breast cancer brain metastasis patients revealed associations of TLR1 and IL6 with poor overall survival. Finally, our finding that a single LPS application at the onset of colonization shapes the later microglia/macrophage reaction at the macro-metastasis brain-parenchyma interface (MMPI) and reduces metastatic infiltration into the brain parenchyma may prove useful in immunotherapeutic considerations.

Identifying Actionable Variants in Cancer - The Dual Web and Batch Processing Tool MTB-Report.

Next-generation sequencing methods continuously provide clinicians and researchers in precision oncology with growing numbers of genomic variants found in cancer. However, manually interpreting the list of variants to identify reliable targets is an inefficient and cumbersome process that does not scale with the increasing number of cases. Support by computer systems is needed for the analysis of large scale experiments and clinical studies to identify new targets and therapies, and user-friendly applications are needed in molecular tumor boards to support clinicians in their decision-making processes. The MTB-Report tool annotates, filters and sorts genetic variants with information from public databases, providing evidence on actionable variants in both scenarios. A web interface supports medical doctors in the tumor board, and a command line mode allows batch processing of large datasets. The MTB-Report tool is available as an R implementation as well as a Docker image to provide a tool that runs out-of-the-box. Moreover, containerization ensures a stable application that delivers reproducible results over time. A public version of the web interface is available at: http://mtb.bioinf.med.uni-goettingen.de/mtb-report.

High-Throughput Profiling of Colorectal Cancer Liver Metastases Reveals Intra- and Inter-Patient Heterogeneity in the EGFR and WNT Pathways Associated with Clinical Outcome.

Seventy percent of patients with colorectal cancer develop liver metastases (CRLM), which are a decisive factor in cancer progression. Therapy outcome is largely influenced by tumor heterogeneity, but the intra- and inter-patient heterogeneity of CRLM has been poorly studied. In particular, the contribution of the WNT and EGFR pathways, which are both frequently deregulated in colorectal cancer, has not yet been addressed in this context. To this end, we comprehensively characterized normal liver tissue and eight CRLM from two patients by standardized histopathological, molecular, and proteomic subtyping. Suitable fresh-frozen tissue samples were profiled by transcriptome sequencing (RNA-Seq) and proteomic profiling with reverse phase protein arrays (RPPA) combined with bioinformatic analyses to assess tumor heterogeneity and identify WNT- and EGFR-related master regulators and metastatic effectors. A standardized data analysis pipeline for integrating RNA-Seq with clinical, proteomic, and genetic data was established. Dimensionality reduction of the transcriptome data revealed a distinct signature for CRLM differing from normal liver tissue and indicated a high degree of tumor heterogeneity. WNT and EGFR signaling were highly active in CRLM and the genes of both pathways were heterogeneously expressed between the two patients as well as between the synchronous metastases of a single patient. An analysis of the master regulators and metastatic effectors implicated in the regulation of these genes revealed a set of four genes (SFN, IGF2BP1, STAT1, PIK3CG) that were differentially expressed in CRLM and were associated with clinical outcome in a large cohort of colorectal cancer patients as well as CRLM samples. In conclusion, high-throughput profiling enabled us to define a CRLM-specific signature and revealed the genes of the WNT and EGFR pathways associated with inter- and intra-patient heterogeneity, which were validated as prognostic biomarkers in CRC primary tumors as well as liver metastases.

Gene-expression profiles of pretreatment biopsies predict complete response of rectal cancer patients to preoperative chemoradiotherapy.

PURPOSE: Preoperative (neoadjuvant) chemoradiotherapy (CRT) and total mesorectal excision is the standard treatment for rectal cancer patients (UICC stage II/III). Up to one-third of patients treated with CRT achieve a pathological complete response (pCR). These patients could be spared from surgery and its associated morbidity and mortality, and assigned to a "watch and wait" strategy. However, reliably identifying pCR based on clinical or imaging parameters remains challenging. EXPERIMENTAL DESIGN: We generated gene-expression profiles of 175 patients with locally advanced rectal cancer enrolled in the CAO/ARO/AIO-94 and -04 trials. One hundred and sixty-one samples were used for building, training and validating a predictor of pCR using a machine learning algorithm. The performance of the classifier was validated in three independent cohorts, comprising 76 patients from (i) the CAO/ARO/AIO-94 and -04 trials (n = 14), (ii) a publicly available dataset (n = 38) and (iii) in 24 prospectively collected samples from the TransValid A trial. RESULTS: A 21-transcript signature yielded the best classification of pCR in 161 patients (Sensitivity: 0.31; AUC: 0.81), when not allowing misclassification of non-complete-responders (False-positive rate = 0). The classifier remained robust when applied to three independent datasets (n = 76). CONCLUSION: The classifier can identify >1/3 of rectal cancer patients with a pCR while never classifying patients with an incomplete response as having pCR. Importantly, we could validate this finding in three independent datasets, including a prospectively collected cohort. Therefore, this classifier could help select rectal cancer patients for a "watch and wait" strategy. TRANSLATIONAL RELEVANCE: Forgoing surgery with its associated side effects could be an option for rectal cancer patients if the prediction of a pathological complete response (pCR) after preoperative chemoradiotherapy would be possible. Based on gene-expression profiles of 161 patients a classifier was developed and validated in three independent datasets (n = 76), identifying over 1/3 of patients with pCR, while never misclassifying a non-complete-responder. Therefore, the classifier can identify patients suited for "watch and wait".

WNT11/ROR2 signaling is associated with tumor invasion and poor survival in breast cancer.

BACKGROUND: Breast cancer has been associated with activation of the WNT signaling pathway, although no driver mutations in WNT genes have been found yet. Instead, a high expression of the alternative WNT receptor ROR2 was observed, in particular in breast cancer brain metastases. However, its respective ligand and downstream signaling in this context remained unknown. METHODS: We modulated the expression of ROR2 in human breast cancer cells and characterized their gene and protein expression by RNA-Seq, qRT-PCR, immunoblots and reverse phase protein array (RPPA) combined with network analyses to understand the molecular basis of ROR2 signaling in breast cancer. Using co-immunoprecipitations, we verified the interaction of ROR2 with the identified ligand, WNT11. The functional consequences of WNT11/ROR2 signaling for tumor cell aggressiveness were assessed by microscopy, impedance sensing as well as viability and invasion assays. To evaluate the translational significance of our findings, we performed gene set enrichment, expression and survival analyses on human breast cancer brain metastases. RESULTS: We found ROR2 to be highly expressed in aggressive breast tumors and associated with worse metastasis-free survival. ROR2 overexpression induced a BRCAness-like phenotype in a cell-context specific manner and rendered cells resistant to PARP inhibition. High levels of ROR2 were furthermore associated with defects in cell morphology and cell-cell-contacts leading to increased tumor invasiveness. On a molecular level, ROR2 overexpression upregulated several non-canonical WNT ligands, in particular WNT11. Co-immunoprecipitation confirmed that WNT11 indeed interacts with the cysteine-rich domain of ROR2 and triggers its invasion-promoting signaling via RHO/ROCK. Knockdown of WNT11 reversed the pro-invasive phenotype and the cellular changes in ROR2-overexpressing cells. CONCLUSIONS: Taken together, our study revealed a novel auto-stimulatory loop in which ROR2 triggers the expression of its own ligand, WNT11, resulting in enhanced tumor invasion associated with breast cancer metastasis.

Coordinated regulation of WNT/ß-catenin, c-Met, and integrin signalling pathways by miR-193b controls triple negative breast cancer metastatic traits.

BACKGROUND: Triple negative breast cancer (TNBC) is the most aggressive subtype of breast cancer (BC). Treatment options for TNBC patients are limited and further insights into disease aetiology are needed to develop better therapeutic approaches. microRNAs' ability to regulate multiple targets could hold a promising discovery approach to pathways relevant for TNBC aggressiveness. Thus, we address the role of miRNAs in controlling three signalling pathways relevant to the biology of TNBC, and their downstream phenotypes. METHODS: To identify miRNAs regulating WNT/ß-catenin, c-Met, and integrin signalling pathways, we performed a high-throughput targeted proteomic approach, investigating the effect of 800 miRNAs on the expression of 62 proteins in the MDA-MB-231 TNBC cell line. We then developed a novel network analysis, Pathway Coregulatory (PC) score, to detect miRNAs regulating these three pathways. Using in vitro assays for cell growth, migration, apoptosis, and stem-cell content, we validated the function of candidate miRNAs. Bioinformatic analyses using BC patients' datasets were employed to assess expression of miRNAs as well as their pathological relevance in TNBC patients. RESULTS: We identified six candidate miRNAs coordinately regulating the three signalling pathways. Quantifying cell growth of three TNBC cell lines upon miRNA gain-of-function experiments, we characterised miR-193b as a strong and consistent repressor of proliferation. Importantly, the effects of miR-193b were stronger than chemical inhibition of the individual pathways. We further demonstrated that miR-193b induced apoptosis, repressed migration, and regulated stem-cell markers in MDA-MB-231 cells. Furthermore, miR-193b expression was the lowest in patients classified as TNBC or Basal compared to other subtypes. Gene Set Enrichment Analysis showed that miR-193b expression was significantly associated with reduced activity of WNT/ß-catenin and c-Met signalling pathways in TNBC patients. CONCLUSIONS: Integrating miRNA-mediated effects and protein functions on networks, we show that miRNAs predominantly act in a coordinated fashion to activate or repress connected signalling pathways responsible for metastatic traits in TNBC. We further demonstrate that our top candidate, miR-193b, regulates these phenotypes to an extent stronger than individual pathway inhibition, thus emphasizing that its effect on TNBC aggressiveness is mediated by the coordinated repression of these functionally interconnected pathways.

Text mining-based word representations for biomedical data analysis and protein-protein interaction networks in machine learning tasks.

Biomedical and life science literature is an essential way to publish experimental results. With the rapid growth of the number of new publications, the amount of scientific knowledge represented in free text is increasing remarkably. There has been much interest in developing techniques that can extract this knowledge and make it accessible to aid scientists in discovering new relationships between biological entities and answering biological questions. Making use of the word2vec approach, we generated word vector representations based on a corpus consisting of over 16 million PubMed abstracts. We developed a text mining pipeline to produce word2vec embeddings with different properties and performed validation experiments to assess their utility for biomedical analysis. An important pre-processing step consisted in the substitution of synonymous terms by their preferred terms in biomedical databases. Furthermore, we extracted gene-gene networks from two embedding versions and used them as prior knowledge to train Graph-Convolutional Neural Networks (CNNs) on large breast cancer gene expression data and on other cancer datasets. Performances of resulting models were compared to Graph-CNNs trained with protein-protein interaction (PPI) networks or with networks derived using other word embedding algorithms. We also assessed the effect of corpus size on the variability of word representations. Finally, we created a web service with a graphical and a RESTful interface to extract and explore relations between biomedical terms using annotated embeddings. Comparisons to biological databases showed that relations between entities such as known PPIs, signaling pathways and cellular functions, or narrower disease ontology groups correlated with higher cosine similarity. Graph-CNNs trained with word2vec-embedding-derived networks performed sufficiently good for the metastatic event prediction tasks compared to other networks. Such performance was good enough to validate the utility of our generated word embeddings in constructing biological networks. Word representations as produced by text mining algorithms like word2vec, therefore are able to capture biologically meaningful relations between entities. Our generated embeddings are publicly available at https://github.com/genexplain/Word2vec-based-Networks/blob/main/README.md.

Predicting the Effect of Variants of Unknown Significance in Molecular Tumor Boards with the VUS-Predict Pipeline.

Precision oncology utilizing molecular biomarkers for targeted therapies is one of the hopes to treat cancer. The availability of patient specific molecular profiling through next-generation sequencing, though, increases the amount of available data per patient to an extent that computational support is required to identify potential driver alterations for targeted therapies and rational decision-making in molecular tumor boards (MTBs). For some genetic variants evidence-based drug recommendations are available in public databases, but for the majority, the variants of unknown significance (VUS), this clinical information is missing. Additionally, for most of these variants no information about the functional impact on the protein is accessible. To acquire maximal functional evidence for VUS, the VUS-Predict pipeline collects estimations about the effect of a VUS by integrating multiple pre-existing tools. Pre-existing tools implement different approaches for their predictions, which are summarized by our newly developed tool with a common score and classification in neutral or deleterious variants. The primary tools are chosen based on their sensitivity and specificity on well-known variants of the transcription factor TP53. Resulting negative and positive predictive values are used to calibrate the VUS-Predict pipeline. Further, the pipeline is evaluated using data from public cancer databases and cases of the MTB in Göttingen, both also in comparison with the ensemble method REVEL. The results show that VUS-Predict has clear advantages in a clinical setting due to clear and traceable predictions. In particular, VUS outperforms REVEL in the real-life setting of a MTB. Likewise, an evaluation on variants of public cancer databases confirms the good results of VUS-Predict and shows the need for a reliable gold standard and unambiguous results of the tools under test.

IGFBP2 Is a Potential Master Regulator Driving the Dysregulated Gene Network Responsible for Short Survival in Glioblastoma Multiforme.

Only 2% of glioblastoma multiforme (GBM) patients respond to standard therapy and survive beyond 36 months (long-term survivors, LTS), while the majority survive less than 12 months (short-term survivors, STS). To understand the mechanism leading to poor survival, we analyzed publicly available datasets of 113 STS and 58 LTS. This analysis revealed 198 differentially expressed genes (DEGs) that characterize aggressive tumor growth and may be responsible for the poor prognosis. These genes belong largely to the Gene Ontology (GO) categories "epithelial-to-mesenchymal transition" and "response to hypoxia." In this article, we applied an upstream analysis approach that involves state-of-the-art promoter analysis and network analysis of the dysregulated genes potentially responsible for short survival in GBM. Binding sites for transcription factors (TFs) associated with GBM pathology like NANOG, NF-κB, REST, FRA-1, PPARG, and seven others were found enriched in the promoters of the dysregulated genes. We reconstructed the gene regulatory network with several positive feedback loops controlled by five master regulators [insulin-like growth factor binding protein 2 (IGFBP2), vascular endothelial growth factor A (VEGFA), VEGF165, platelet-derived growth factor A (PDGFA), adipocyte enhancer-binding protein (AEBP1), and oncostatin M (OSMR)], which can be proposed as biomarkers and as therapeutic targets for enhancing GBM prognosis. A critical analysis of this gene regulatory network gives insights into the mechanism of gene regulation by IGFBP2 via several TFs including the key molecule of GBM tumor invasiveness and progression, FRA-1. All the observations were validated in independent cohorts, and their impact on overall survival has been investigated.

Effects of the Novel PFKFB3 Inhibitor KAN0438757 on Colorectal Cancer Cells and Its Systemic Toxicity Evaluation In Vivo.

BACKGROUND: Despite substantial progress made in the last decades in colorectal cancer (CRC) research, new treatment approaches are still needed to improve patients' long-term survival. To date, the promising strategy to target tumor angiogenesis metabolically together with a sensitization of CRC to chemo- and/or radiotherapy by PFKFB3 (6-phosphofructo-2-kinase/fructose-2,6-biphosphatase-3) inhibition has never been tested. Therefore, initial evaluation and validation of newly developed compounds such as KAN0438757 and their effects on CRC cells are crucial steps preceding to in vivo preclinical studies, which in turn may consolidate new therapeutic targets. MATERIALS AND METHODS: The efficiency of KAN0438757 to block PFKFB3 expression and translation in human CRC cells was evaluated by immunoblotting and real-time PCR. Functional in vitro assays assessed the effects of KAN0438757 on cell viability, proliferation, survival, adhesion, migration and invasion. Additionally, we evaluated the effects of KAN0438757 on matched patient-derived normal and tumor organoids and its systemic toxicity in vivo in C57BL6/N mice. RESULTS: High PFKFB3 expression is correlated with a worse survival in CRC patients. KAN0438757 reduces PFKFB3 protein expression without affecting its transcriptional regulation. Additionally, a concentration-dependent anti-proliferative effect was observed. The migration and invasion capacity of cancer cells were significantly reduced, independent of the anti-proliferative effect. When treating colonic patient-derived organoids with KAN0438757 an impressive effect on tumor organoids growth was apparent, surprisingly sparing normal colonic organoids. No high-grade toxicity was observed in vivo. CONCLUSION: The PFKFB3 inhibitor KAN0438757 significantly reduced CRC cell migration, invasion and survival. Moreover, on patient-derived cancer organoids KAN0438757 showed significant effects on growth, without being overly toxic in normal colon organoids and healthy mice. Our findings strongly encourage further translational studies to evaluate KAN0438757 in CRC therapy.

Explaining decisions of graph convolutional neural networks: patient-specific molecular subnetworks responsible for metastasis prediction in breast cancer.

BACKGROUND: Contemporary deep learning approaches show cutting-edge performance in a variety of complex prediction tasks. Nonetheless, the application of deep learning in healthcare remains limited since deep learning methods are often considered as non-interpretable black-box models. However, the machine learning community made recent elaborations on interpretability methods explaining data point-specific decisions of deep learning techniques. We believe that such explanations can assist the need in personalized precision medicine decisions via explaining patient-specific predictions. METHODS: Layer-wise Relevance Propagation (LRP) is a technique to explain decisions of deep learning methods. It is widely used to interpret Convolutional Neural Networks (CNNs) applied on image data. Recently, CNNs started to extend towards non-Euclidean domains like graphs. Molecular networks are commonly represented as graphs detailing interactions between molecules. Gene expression data can be assigned to the vertices of these graphs. In other words, gene expression data can be structured by utilizing molecular network information as prior knowledge. Graph-CNNs can be applied to structured gene expression data, for example, to predict metastatic events in breast cancer. Therefore, there is a need for explanations showing which part of a molecular network is relevant for predicting an event, e.g., distant metastasis in cancer, for each individual patient. RESULTS: We extended the procedure of LRP to make it available for Graph-CNN and tested its applicability on a large breast cancer dataset. We present Graph Layer-wise Relevance Propagation (GLRP) as a new method to explain the decisions made by Graph-CNNs. We demonstrate a sanity check of the developed GLRP on a hand-written digits dataset and then apply the method on gene expression data. We show that GLRP provides patient-specific molecular subnetworks that largely agree with clinical knowledge and identify common as well as novel, and potentially druggable, drivers of tumor progression. CONCLUSIONS: The developed method could be potentially highly useful on interpreting classification results in the context of different omics data and prior knowledge molecular networks on the individual patient level, as for example in precision medicine approaches or a molecular tumor board.