Abnormality Detection and Failure Prediction Using Explainable Bayesian Deep Learning: Methodology and Case Study of Real-World Gas Turbine Anomalies

: Mistrust, amplified by numerous artificial intelligence (AI) related incidents, has caused the energy and industrial sectors to be amongst the slowest adopter of AI methods. Central to this issue is the black-box problem of AI, which impedes investments and fast becoming a legal hazard for users. Explainable AI (XAI) is a recent paradigm to tackle this challenge. Being the backbone of the industry, the prognostic and health management (PHM) domain has recently been introduced to XAI. However, many deficiencies, particularly lack of explanation assessment methods and uncertainty quantification, plague this young field. In this paper, we elaborate a framework on explainable anomaly detection and failure prognostic employing a Bayesian deep learning model to generate local and global explanations from the PHM tasks. An uncertainty measure of the Bayesian model is utilized as marker for anomalies expanding the prognostic explanation scope to include model’s confidence. Also, the global explanation is used to improve prognostic performance, an aspect neglected from the handful of PHM-XAI publications. The quality of the explanation is finally examined employing local accuracy and consistency properties. The method is tested on real-world gas turbine anomalies and synthetic turbofan data failure prediction. Seven out of eight of the tested anomalies were successfully identified. Additionally, the prognostic outcome showed 19% improvement in statistical terms and achieved the highest prognostic score amongst best published results on the topic.

Exome sequencing of glioblastoma-derived cancer stem cells reveals rare clinically relevant frameshift deletion in MLLT1 gene

Background Glioblastoma multiforme (GBM) is a heterogeneous CNS neoplasm which causes significant morbidity and mortality. One reason for the poor prognostic outcome of GBM is attributed to the presence of cancer stem cells (CSC) which confer resistance against standard chemo- and radiotherapeutics modalities. Two types of GBM-associated CSC were isolated from the same patient: tumor core- (c-CSC) and peritumor tissue-derived cancer stem cells (p-CSC). Our experiments are focused on glioblastoma–IDH-wild type, and no disease-defining alterations were present in histone, BRAF or other genes. Methods In the present study, potential differences in genetic variants between c-CSC versus p-CSC derived from four GBM patients were investigated with the aims of (1) comparing the exome sequences between all the c-CSC or p-CSC to identify the common variants; (2) identifying the variants affecting the function of genes known to be involved in cancer origin and development. Results By comparative analyses, we identified common gene single nucleotide variants (SNV) in all GBM c-CSC and p-CSC, a potentially deleterious variant was a frameshift deletion at Gln461fs in the MLLT1 gene, that was encountered only in p-CSC samples with different allelic frequency. Conclusions We discovered a potentially harmful frameshift deletion at Gln461fs in the MLLT1 gene. Further investigation is required to confirm the presence of the identified mutations in patient tissue samples, as well as the significance of the frameshift mutation in the MLLT1 gene on GBM biology and response to therapy based on genomic functional experiments.

Knowledge, opinions, and practices related to oral cancer prevention and oral mucosal examination among dentists in Moldova, Belarus and Armenia: a multi-country cross-sectional study

Introduction Moldova, Belarus, and Armenia are post-Soviet countries with a high rate of heavy smokers and a relatively high age-standardized incidence of oral cancer. However, to our knowledge, there is lack of available information on dentists’ knowledge on prevention of oral cancer in the countries in question. Accordingly, this study aimed to assess the knowledge, opinions, and practices related to oral cancer prevention and oral mucosal examination among dentists in Moldova, Belarus, and Armenia. Methods This was a multi-country, cross-sectional study based on a self-administered questionnaire. A structured questionnaire was distributed to 3534 dentists (797 in Chisinau, Moldova, 1349 in Minsk, Belarus, and 1388 in Yerevan, Armenia). Dentists' knowledge about risk factors for oral cancer development and its clinical picture, current practices and opinions with regard to oral mucosal screening and oral cancer prevention, and their consistency to perform oral mucosal examination were assessed. A knowledge score ranging from 0 to 14 points was generated based on each dentist’s answer to the questionnaire. Results A total of 1316 dentists responded, achieving an overall response rate of 37.2% (34.5% in Moldova; 52.3% in Belarus; 24.2% in Armenia). Most dentists in the three countries correctly identified tobacco (83.8–98.2%) and prior oral cancer lesions (84.0–96.3%) as risk factors for oral cancer. Most dentists correctly identified leukoplakia as a lesion with malignant potential (68.7% in Moldova; 88.5% in Belarus; 69.9% in Armenia), while erythroplakia was identified by much fewer in all three countries. Less than 52% of dentists identified the tongue, rim of tongue, and floor of mouth as the most common sites for oral cancer. The mean knowledge score for all countries combined was 7.5 ± 2.7. The most commonly reported barriers to perform oral mucosal examination were lack of training, knowledge, and experience. Conclusions This study highlights the need for improved oral cancer-related education and training on oral mucosal examination for dentists in Moldova, Belarus, and Armenia. Such skills are essential to enhance oral cancer prevention and to improve the prognostic outcome by early detection.

m5C Regulator-Mediated Methylation Modification Patterns and Tumor Microenvironment Infiltration Characterization in Papillary Thyroid Carcinoma

Recently, immune response modulation at the epigenetic level is illustrated in studies, but the possible function of RNA 5-methylcytosine (m5C) modification in cell infiltration within the tumor microenvironment (TME) is still unclear. Three different m5C modification patterns were identified, and high differentiation degree was observed in the cell infiltration features within TME under the above three identified patterns. A low m5C-score, which was reflected in the activated immunity, predicted the relatively favorable prognostic outcome. A small amount of effective immune infiltration was seen in the high m5C-score subtype, indicating the dismal patient survival. Our study constructed a diagnostic model using the 10 signature genes highly related to the m5C-score, discovered that the model exhibited high diagnostic accuracy for PTC, and screened out five potential drugs for PTC based on this m5C-score model. m5C modification exerts an important part in forming the TME complexity and diversity. It is valuable to evaluate the m5C modification patterns in single tumors, so as to enhance our understanding towards the infiltration characterization in TME.

Identification of N6-Methyladenosine-Related LncRNAs for Predicting Overall Survival and Clustering of a Potentially Novel Molecular Subtype of Breast Cancer

We aimed to identify a signature comprising N6-methyladenosine (m6A)-related long non-coding RNAs (lncRNAs) and molecular subtypes associated with breast cancer (BRCA). We obtained data of BRCA samples from The Cancer Genome Atlas database. The m6A-related lncRNA prognostic signature (m6A-LPS) included 10 lncRNAs previously identified as prognostic m6A-related lncRNAs and was constructed using integrated bioinformatics analysis and validated. Accordingly, a risk score based on the m6A-LPS signature was established and shown to confirm differences in survival between high-risk and low-risk groups. Three distinct genotypes were identified, whose characteristics included features of the tumor immune microenvironment in each subtype. Our results indicated that patients in Cluster 2 might have a worse prognostic outcome than those in other clusters. The three genotypes and risk subgroups were enriched in different biological processes and pathways, respectively. We then constructed a competing endogenous RNA network based on the prognostic m6A-related lncRNAs. Finally, we validated the expression levels of target lncRNAs in 72 clinical samples. In summary, the m6A-LPS and the potentially novel genotype may provide a theoretical basis for further study of the molecular mechanism of BRCA and may provide novel insights into precision medicine.

Digitalis Therapy in Patients With Ventricular Tachyarrhythmias

Objective: The study sought to assess the prognostic value of treatment with digitalis on long-term prognosis in patients with ventricular tachyarrhythmias and atrial fibrillation (AF) and/or heart failure (HF).Background: Data regarding outcome of digitalis therapy following ventricular tachyarrhythmias is limited.Methods: A large retrospective registry was used including consecutive patients with episodes of ventricular tachycardia (VT) or fibrillation (VF) from 2002 to 2015. Patients treated with digitalis were compared to patients without. The primary prognostic outcome was all-cause mortality at three years, secondary endpoints comprised of a composite arrhythmic endpoint (i.e., recurrences of ventricular tachyarrhythmias, sudden cardiac death) and cardiac rehospitalization. Kaplan Meier, multivariable cox regression and time trend analyses were applied for statistics.Results: A total of 831 patients were included (20% treated with digitalis and 80% without). At three years, digitalis treatment was not associated with all-cause mortality in patients with ventricular tachyarrhythmias (24% vs. 21%, log rank p=0.736; HR=1.063; 95% CI 0.746-1.515; p=0.736). However, digitalis therapy was associated with increased risk of the composite endpoint (38% vs. 23%; log rank p=0.001; HR=1.719; 95% CI 1.279-2.311; p=0.001) and cardiac rehospitalization (31% vs. 18%; log rank p=0.001; HR=1.829; 95% CI 1.318-2.538; p=0.001) at three years, which was still evident within multivariable Cox regression analyses. Finally, digitoxin was associated with worse prognosis than digoxin.Conclusion: Digitalis therapy was not associated mortality in patients with ventricular tachyarrhythmias, but with increased risk of the composite arrhythmic endpoint and cardiac rehospitalization at three years.

Exome Sequencing of Glioblastoma-Derived Cancer Stem Cells Reveals Rare Clinically Relevant Frameshift Deletion in MLLT1 Gene

Background: Glioblastoma multiforme (GBM) is a heterogeneous CNS neoplasm which causes significant morbidity and mortality. One reason for the poor prognostic outcome of GBM is attributed to the presence of cancer stem cells (CSC) which confer resistance against standard chemo- and radiotherapeutics modalities. Two types of GBM-associated CSC were isolated from the same patient: tumor core- (c-CSC) and peritumor tissue-derived cancer stem cells (p-CSC).Methods: In the present study, potential differences in genetic variants between c-CSC versus p-CSC derived from four GBM patients were investigated with the aims of 1) comparing the exome sequences between all the c-CSC or p-CSC to identify the common variants; 2) identifying the variants affecting the function of genes known to be involved in cancer origin and development.Results: By comparative analyses, we identified common gene single nucleotide variants (SNV) in all GBM c-CSC and p-CSC, a potentially deleterious variant was a frameshift deletion at Gln461fs in the MLLT1 gene, that was encountered only in p-CSC samples with different allelic frequency.Conclusions: Our study supports the hypothesis that the varied genetic composition of GBM-associated c-CSC and p-CSC may be involved in different therapeutic responses or the recurrent nature of GBM.