Mutational signature-based identification of DNA repair deficient gastroesophageal adenocarcinomas for therapeutic targeting.

Homologous recombination (HR) and nucleotide excision repair (NER) are the two most frequently disabled DNA repair pathways in cancer. HR-deficient breast, ovarian, pancreatic and prostate cancers respond well to platinum chemotherapy and PARP inhibitors. However, the frequency of HR deficiency in gastric and esophageal adenocarcinoma (GEA) still lacks diagnostic and functional validation. Using whole exome and genome sequencing data, we found that a significant subset of GEA, but very few colorectal adenocarcinomas, show evidence of HR deficiency by mutational signature analysis (HRD score). High HRD gastric cancer cell lines demonstrated functional HR deficiency by RAD51 foci assay and increased sensitivity to platinum chemotherapy and PARP inhibitors. Of clinical relevance, analysis of three different GEA patient cohorts demonstrated that platinum treated HR deficient cancers had better outcomes. A gastric cancer cell line with strong sensitivity to cisplatin showed HR proficiency but exhibited NER deficiency by two photoproduct repair assays. Single-cell RNA-sequencing revealed that, in addition to inducing apoptosis, cisplatin treatment triggered ferroptosis in a NER-deficient gastric cancer, validated by intracellular GSH assay. Overall, our study provides preclinical evidence that a subset of GEAs harbor genomic features of HR and NER deficiency and may therefore benefit from platinum chemotherapy and PARP inhibitors.

Increased frequency of CHD1 deletions in prostate cancers of African American men is associated with rapid disease progression without inducing homologous recombination deficiency.

We analyzed genomic data derived from the prostate cancer of African and European American men in order to identify differences that may contribute to racial disparity of outcome and that could also define novel therapeutic strategies. In addition to analyzing patient derived next generation sequencing data, we performed FISH based confirmatory studies of Chromodomain helicase DNA-binding protein 1 (CHD1) loss on prostate cancer tissue microarrays. We created CRISPR edited, CHD1 deficient prostate cancer cell lines for genomic, drug sensitivity and functional homologous recombination (HR) activity analysis. We found that subclonal deletion of CHD1 is nearly three times as frequent in prostate tumors of African American men than in men of European ancestry and it associates with rapid disease progression. We further showed that CHD1 deletion is not associated with homologous recombination deficiency associated mutational signatures in prostate cancer. In prostate cancer cell line models CHD1 deletion did not induce HR deficiency as detected by RAD51 foci formation assay or mutational signatures, which was consistent with the moderate increase of olaparib sensitivity. CHD1 deficient prostate cancer cells, however, showed higher sensitivity to talazoparib. CHD1 loss may contribute to worse outcome of prostate cancer in African American men. A deeper understanding of the interaction between CHD1 loss and PARP inhibitor sensitivity will be needed to determine the optimal use of targeted agents such as talazoparib in the context of castration resistant prostate cancer.

Mobilisation and analyses of publicly available SARS-CoV-2 data for pandemic responses.

The COVID-19 pandemic has seen large-scale pathogen genomic sequencing efforts, becoming part of the toolbox for surveillance and epidemic research. This resulted in an unprecedented level of data sharing to open repositories, which has actively supported the identification of SARS-CoV-2 structure, molecular interactions, mutations and variants, and facilitated vaccine development and drug reuse studies and design. The European COVID-19 Data Platform was launched to support this data sharing, and has resulted in the deposition of several million SARS-CoV-2 raw reads. In this paper we describe (1) open data sharing, (2) tools for submission, analysis, visualisation and data claiming (e.g. ORCiD), (3) the systematic analysis of these datasets, at scale via the SARS-CoV-2 Data Hubs as well as (4) lessons learnt. This paper describes a component of the Platform, the SARS-CoV-2 Data Hubs, which enable the extension and set up of infrastructure that we intend to use more widely in the future for pathogen surveillance and pandemic preparedness.

Annotated dataset for training deep learning models to detect astrocytes in human brain tissue.

Astrocytes, a type of glial cell, significantly influence neuronal function, with variations in morphology and density linked to neurological disorders. Traditional methods for their accurate detection and density measurement are laborious and unsuited for large-scale operations. We introduce a dataset from human brain tissues stained with aldehyde dehydrogenase 1 family member L1 (ALDH1L1) and glial fibrillary acidic protein (GFAP). The digital whole slide images of these tissues were partitioned into 8730 patches of 500 × 500 pixels, comprising 2323 ALDH1L1 and 4714 GFAP patches at a pixel size of 0.5019/pixel, furthermore 1382 ADHD1L1 and 311 GFAP patches at 0.3557/pixel. Sourced from 16 slides and 8 patients our dataset promotes the development of tools for glial cell detection and quantification, offering insights into their density distribution in various brain areas, thereby broadening neuropathological study horizons. These samples hold value for automating detection methods, including deep learning. Derived from human samples, our dataset provides a platform for exploring astrocyte functionality, potentially guiding new diagnostic and treatment strategies for neurological disorders.

Systematic detection of co-infection and intra-host recombination in more than 2 million global SARS-CoV-2 samples.

Systematic monitoring of SARS-CoV-2 co-infections between different lineages and assessing the risk of intra-host recombinant emergence are crucial for forecasting viral evolution. Here we present a comprehensive analysis of more than 2 million SARS-CoV-2 raw read datasets submitted to the European COVID-19 Data Portal to identify co-infections and intra-host recombination. Co-infection was observed in 0.35% of the investigated cases. Two independent procedures were implemented to detect intra-host recombination. We show that sensitivity is predominantly determined by the density of lineage-defining mutations along the genome, thus we used an expanded list of mutually exclusive defining mutations of specific variant combinations to increase statistical power. We call attention to multiple challenges rendering recombinant detection difficult and provide guidelines for the reduction of false positives arising from chimeric sequences produced during PCR amplification. Additionally, we identify three recombination hotspots of Delta - Omicron BA.1 intra-host recombinants.

Biologically informed deep learning for explainable epigenetic clocks.

Ageing is often characterised by progressive accumulation of damage, and it is one of the most important risk factors for chronic disease development. Epigenetic mechanisms including DNA methylation could functionally contribute to organismal aging, however the key functions and biological processes may govern ageing are still not understood. Although age predictors called epigenetic clocks can accurately estimate the biological age of an individual based on cellular DNA methylation, their models have limited ability to explain the prediction algorithm behind and underlying key biological processes controlling ageing. Here we present XAI-AGE, a biologically informed, explainable deep neural network model for accurate biological age prediction across multiple tissue types. We show that XAI-AGE outperforms the first-generation age predictors and achieves similar results to deep learning-based models, while opening up the possibility to infer biologically meaningful insights of the activity of pathways and other abstract biological processes directly from the model.

Nucleotide excision repair deficiency is a targetable therapeutic vulnerability in clear cell renal cell carcinoma.

Due to a demonstrated lack of DNA repair deficiencies, clear cell renal cell carcinoma (ccRCC) has not benefitted from targeted synthetic lethality-based therapies. We investigated whether nucleotide excision repair (NER) deficiency is present in an identifiable subset of ccRCC cases that would render those tumors sensitive to therapy targeting this specific DNA repair pathway aberration. We used functional assays that detect UV-induced 6-4 pyrimidine-pyrimidone photoproducts to quantify NER deficiency in ccRCC cell lines. We also measured sensitivity to irofulven, an experimental cancer therapeutic agent that specifically targets cells with inactivated transcription-coupled nucleotide excision repair (TC-NER). In order to detect NER deficiency in clinical biopsies, we assessed whole exome sequencing data for the presence of an NER deficiency associated mutational signature previously identified in ERCC2 mutant bladder cancer. Functional assays showed NER deficiency in ccRCC cells. Some cell lines showed irofulven sensitivity at a concentration that is well tolerated by patients. Prostaglandin reductase 1 (PTGR1), which activates irofulven, was also associated with this sensitivity. Next generation sequencing data of the cell lines showed NER deficiency-associated mutational signatures. A significant subset of ccRCC patients had the same signature and high PTGR1 expression. ccRCC cell line-based analysis showed that NER deficiency is likely present in this cancer type. Approximately 10% of ccRCC patients in the TCGA cohort showed mutational signatures consistent with ERCC2 inactivation associated NER deficiency and also substantial levels of PTGR1 expression. These patients may be responsive to irofulven, a previously abandoned anticancer agent that has minimal activity in NER-proficient cells.

Bacterial colony size growth estimation by deep learning.

The bacterial growth rate is important for pathogenicity and food safety. Therefore, the study of bacterial growth rate over time can provide important data from a medical and veterinary point of view. We trained convolutional neural networks (CNNs) on manually annotated solid medium cultures to detect bacterial colonies as accurately as possible. Predictions of bacterial colony size and growth rate were estimated from image sequences of independent Staphylococcus aureus cultures using trained CNNs. A simple linear model for control cultures with less than 150 colonies estimated that the mean growth rate was 60.3 [Formula: see text] for the first 24 h. Analyzing with a mixed effect model that also takes into account the effect of culture, smaller values of change in colony size were obtained (control: 51.0 [Formula: see text], rifampicin pretreated: 36.5[Formula: see text]). An increase in the number of neighboring colonies clearly reduces the colony growth rate in the control group but less typically in the rifampicin-pretreated group. Based on our results, CNN-based bacterial colony detection and the subsequent analysis of bacterial colony growth dynamics might become an accurate and efficient tool for bacteriological work and research.

A biallelic multiple nucleotide length polymorphism explains functional causality at 5p15.33 prostate cancer risk locus.

To date, single-nucleotide polymorphisms (SNPs) have been the most intensively investigated class of polymorphisms in genome wide associations studies (GWAS), however, other classes such as insertion-deletion or multiple nucleotide length polymorphism (MNLPs) may also confer disease risk. Multiple reports have shown that the 5p15.33 prostate cancer risk region is a particularly strong expression quantitative trait locus (eQTL) for Iroquois Homeobox 4 (IRX4) transcripts. Here, we demonstrate using epigenome and genome editing that a biallelic (21 and 47 base pairs (bp)) MNLP is the causal variant regulating IRX4 transcript levels. In LNCaP prostate cancer cells (homozygous for the 21 bp short allele), a single copy knock-in of the 47 bp long allele potently alters the chromatin state, enabling de novo functional binding of the androgen receptor (AR) associated with increased chromatin accessibility, Histone 3 lysine 27 acetylation (H3K27ac), and ~3-fold upregulation of IRX4 expression. We further show that an MNLP is amongst the strongest candidate susceptibility variants at two additional prostate cancer risk loci. We estimated that at least 5% of prostate cancer risk loci could be explained by functional non-SNP causal variants, which may have broader implications for other cancers GWAS. More generally, our results underscore the importance of investigating other classes of inherited variation as causal mediators of human traits.

Annotated dataset for deep-learning-based bacterial colony detection.

Quantifying bacteria per unit mass or volume is a common task in various fields of microbiology (e.g., infectiology and food hygiene). Most bacteria can be grown on culture media. The unicellular bacteria reproduce by dividing into two cells, which increases the number of bacteria in the population. Methodologically, this can be followed by culture procedures, which mostly involve determining the number of bacterial colonies on the solid culture media that are visible to the naked eye. However, it is a time-consuming and laborious professional activity. Addressing the automation of colony counting by convolutional neural networks in our work, we have cultured 24 bacteria species of veterinary importance with different concentrations on solid media. A total of 56,865 colonies were annotated manually by bounding boxes on the 369 digital images of bacterial cultures. The published dataset will help developments that use artificial intelligence to automate the counting of bacterial colonies.

Genomic Landscape of Normal and Breast Cancer Tissues in a Hungarian Pilot Cohort.

A limited number of studies have focused on the mutational landscape of breast cancer in different ethnic populations within Europe and compared the data with other ethnic groups and databases. We performed whole-genome sequencing of 63 samples from 29 Hungarian breast cancer patients. We validated a subset of the identified variants at the DNA level using the Illumina TruSight Oncology (TSO) 500 assay. Canonical breast-cancer-associated genes with pathogenic germline mutations were CHEK2 and ATM. Nearly all the observed germline mutations were as frequent in the Hungarian breast cancer cohort as in independent European populations. The majority of the detected somatic short variants were single-nucleotide polymorphisms (SNPs), and only 8% and 6% of them were deletions or insertions, respectively. The genes most frequently affected by somatic mutations were KMT2C (31%), MUC4 (34%), PIK3CA (18%), and TP53 (34%). Copy number alterations were most common in the NBN, RAD51C, BRIP1, and CDH1 genes. For many samples, the somatic mutational landscape was dominated by mutational processes associated with homologous recombination deficiency (HRD). Our study, as the first breast tumor/normal sequencing study in Hungary, revealed several aspects of the significantly mutated genes and mutational signatures, and some of the copy number variations and somatic fusion events. Multiple signs of HRD were detected, highlighting the value of the comprehensive genomic characterization of breast cancer patient populations.

Liquid biopsy-based monitoring of residual disease in multiple myeloma by analysis of the rearranged immunoglobulin genes-A feasibility study.

The need for sensitive monitoring of minimal/measurable residual disease (MRD) in multiple myeloma emerged as novel therapies led to deeper responses. Moreover, the potential benefits of blood-based analyses, the so-called liquid biopsy is prompting more and more studies to assess its feasibility. Considering these recent demands, we aimed to optimize a highly sensitive molecular system based on the rearranged immunoglobulin (Ig) genes to monitor MRD from peripheral blood. We analyzed a small group of myeloma patients with the high-risk t(4;14) translocation, using next-generation sequencing of Ig genes and droplet digital PCR of patient-specific Ig heavy chain (IgH) sequences. Moreover, well established monitoring methods such as multiparametric flow cytometry and RT-qPCR of the fusion transcript IgH::MMSET (IgH and multiple myeloma SET domain-containing protein) were utilized to evaluate the feasibility of these novel molecular tools. Serum measurements of M-protein and free light chains together with the clinical assessment by the treating physician served as routine clinical data. We found significant correlation between our molecular data and clinical parameters, using Spearman correlations. While the comparisons of the Ig-based methods and the other monitoring methods (flow cytometry, qPCR) were not statistically evaluable, we found common trends in their target detection. Regarding longitudinal disease monitoring, the applied methods yielded complementary information thus increasing the reliability of MRD evaluation. We also detected indications of early relapse before clinical signs, although this implication needs further verification in a larger patient cohort.

SARS-CoV-2 receptor-binding domain deep mutational AlphaFold2 structures.

Leveraging recent advances in computational modeling of proteins with AlphaFold2 (AF2) we provide a complete curated data set of all single mutations from each of the 7 main SARS-CoV-2 lineages spike protein receptor binding domain (RBD) resulting in 3819X7 = 26733 PDB structures. We visualize the generated structures and show that AF2 pLDDT values are correlated with state-of-the-art disorder approximations, implying some internal protein dynamics are also captured by the model. Joint increasing mutational coverage of both structural and phenotype data coupled with advances in machine learning can be leveraged to accelerate virology research, specifically future variant prediction. We hope this data release can offer assistance into further understanding of the local and global mutational landscape of SARS-CoV-2 as well as provide insight into the biological understanding that 3D structure acts as a bridge between protein genotype and phenotype.

Automated prediction of COVID-19 severity upon admission by chest X-ray images and clinical metadata aiming at accuracy and explainability.

In the past few years COVID-19 posed a huge threat to healthcare systems around the world. One of the first waves of the pandemic hit Northern Italy severely resulting in high casualties and in the near breakdown of primary care. Due to these facts, the Covid CXR Hackathon-Artificial Intelligence for Covid-19 prognosis: aiming at accuracy and explainability challenge had been launched at the beginning of February 2022, releasing a new imaging dataset with additional clinical metadata for each accompanying chest X-ray (CXR). In this article we summarize our techniques at correctly diagnosing chest X-ray images collected upon admission for severity of COVID-19 outcome. In addition to X-ray imagery, clinical metadata was provided and the challenge also aimed at creating an explainable model. We created a best-performing, as well as, an explainable model that makes an effort to map clinical metadata to image features whilst predicting the prognosis. We also did many ablation studies in order to identify crucial parts of the models and the predictive power of each feature in the datasets. We conclude that CXRs at admission do not help the predicting power of the metadata significantly by itself and contain mostly information that is also mutually present in the blood samples and other clinical factors collected at admission.

Patterns of Somatic Variants in Colorectal Adenoma and Carcinoma Tissue and Matched Plasma Samples from the Hungarian Oncogenome Program.

Analysis of circulating cell-free DNA (cfDNA) of colorectal adenoma (AD) and cancer (CRC) patients provides a minimally invasive approach that is able to explore genetic alterations. It is unknown whether there are specific genetic variants that could explain the high prevalence of CRC in Hungary. Whole-exome sequencing (WES) was performed on colon tissues (27 AD, 51 CRC) and matched cfDNAs (17 AD, 33 CRC); furthermore, targeted panel sequencing was performed on a subset of cfDNA samples. The most frequently mutated genes were APC, KRAS, and FBN3 in AD, while APC, TP53, TTN, and KRAS were the most frequently mutated in CRC tissue. Variants in KRAS codons 12 (AD: 8/27, CRC: 11/51 (0.216)) and 13 (CRC: 3/51 (0.06)) were the most frequent in our sample set, with G12V (5/27) dominance in ADs and G12D (5/51 (0.098)) in CRCs. In terms of the cfDNA WES results, tumor somatic variants were found in 6/33 of CRC cases. Panel sequencing revealed somatic variants in 8 out of the 12 enrolled patients, identifying 12/20 tumor somatic variants falling on its targeted regions, while WES recovered only 20% in the respective regions in cfDNA of the same patients. In liquid biopsy analyses, WES is less efficient compared to the targeted panel sequencing with a higher coverage depth that can hold a relevant clinical potential to be applied in everyday practice in the future.

Nucleotide excision repair deficiency is a targetable therapeutic vulnerability in clear cell renal cell carcinoma.

Purpose: Due to a demonstrated lack of DNA repair deficiencies, clear cell renal cell carcinoma (ccRCC) has not benefitted from targeted synthetic lethality-based therapies. We investigated whether nucleotide excision repair (NER) deficiency is present in an identifiable subset of ccRCC cases that would render those tumors sensitive to therapy targeting this specific DNA repair pathway aberration. Experimental Design: We used functional assays that detect UV-induced 6-4 pyrimidine-pyrimidone photoproducts to quantify NER deficiency in ccRCC cell lines. We also measured sensitivity to irofulven, an experimental cancer therapeutic agent that specifically targets cells with inactivated transcription-coupled nucleotide excision repair (TC-NER). In order to detect NER deficiency in clinical biopsies, we assessed whole exome sequencing data for the presence of an NER deficiency associated mutational signature previously identified in ERCC2 mutant bladder cancer. Results: Functional assays showed NER deficiency in ccRCC cells. Irofulven sensitivity increased in some cell lines. Prostaglandin reductase 1 (PTGR1), which activates irofulven, was also associated with this sensitivity. Next generation sequencing data of the cell lines showed NER deficiency-associated mutational signatures. A significant subset of ccRCC patients had the same signature and high PTGR1 expression. Conclusions: ccRCC cell line based analysis showed that NER deficiency is likely present in this cancer type. Approximately 10% of ccRCC patients in the TCGA cohort showed mutational signatures consistent with ERCC2 inactivation associated NER deficiency and also substantial levels of PTGR1 expression. These patients may be responsive to irofulven, a previously abandoned anticancer agent that has minimal activity in NER-proficient cells.

Impact Evaluation of Score Classes and Annotation Regions in Deep Learning-Based Dairy Cow Body Condition Prediction.

Body condition scoring is a simple method to estimate the energy supply of dairy cattle. Our study aims to investigate the accuracy with which supervised machine learning, specifically a deep convolutional neural network (CNN), can be used to retrieve body condition score (BCS) classes estimated by an expert. We recorded images of animals' rumps in three large-scale farms using a simple action camera. The images were annotated with classes and three different-sized bounding boxes by an expert. A CNN pretrained model was fine-tuned on 12 and 3 BCS classes. Training in 12 classes with a 0 error range, the Cohen's kappa value yielded minimal agreement between the model predictions and ground truth. Allowing an error range of 0.25, we obtained minimum or weak agreement. With an error range of 0.5, we had strong or almost perfect agreement. The kappa values for the approach trained on three classes show that we can classify all animals into BCS categories with at least moderate agreement. Furthermore, CNNs trained on 3 BCS classes showed a remarkably higher proportion of strong agreement than those trained in 12 classes. The prediction precision when training with various annotation region sizes showed no meaningful differences. The weights of our trained CNNs are freely available, supporting similar works.

Physical imaging parameter variation drives domain shift.

Statistical learning algorithms strongly rely on an oversimplified assumption for optimal performance, that is, source (training) and target (testing) data are independent and identically distributed. Variation in human tissue, physician labeling and physical imaging parameters (PIPs) in the generative process, yield medical image datasets with statistics that render this central assumption false. When deploying models, new examples are often out of distribution with respect to training data, thus, training robust dependable and predictive models is still a challenge in medical imaging with significant accuracy drops common for deployed models. This statistical variation between training and testing data is referred to as domain shift (DS).To the best of our knowledge we provide the first empirical evidence that variation in PIPs between test and train medical image datasets is a significant driver of DS and model generalization error is correlated with this variance. We show significant covariate shift occurs due to a selection bias in sampling from a small area of PIP space for both inter and intra-hospital regimes. In order to show this, we control for population shift, prevalence shift, data selection biases and annotation biases to investigate the sole effect of the physical generation process on model generalization for a proxy task of age group estimation on a combined 44 k image mammogram dataset collected from five hospitals.We hypothesize that training data should be sampled evenly from PIP space to produce the most robust models and hope this study provides motivation to retain medical image generation metadata that is almost always discarded or redacted in open source datasets. This metadata measured with standard international units can provide a universal regularizing anchor between distributions generated across the world for all current and future imaging modalities.