K-PathVQA: Knowledge-Aware Multimodal Representation for Pathology Visual Question Answering.

Pathology imaging is routinely used to detect the underlying effects and causes of diseases or injuries. Pathology visual question answering (PathVQA) aims to enable computers to answer questions about clinical visual findings from pathology images. Prior work on PathVQA has focused on directly analyzing the image content using conventional pretrained encoders without utilizing relevant external information when the image content is inadequate. In this paper, we present a knowledge-driven PathVQA (K-PathVQA), which uses a medical knowledge graph (KG) from a complementary external structured knowledge base to infer answers for the PathVQA task. K-PathVQA improves the question representation with external medical knowledge and then aggregates vision, language, and knowledge embeddings to learn a joint knowledge-image-question representation. Our experiments using a publicly available PathVQA dataset showed that our K-PathVQA outperformed the best baseline method with an increase of 4.15% in accuracy for the overall task, an increase of 4.40% in open-ended question type and an absolute increase of 1.03% in closed-ended question types. Ablation testing shows the impact of each of the contributions. Generalizability of the method is demonstrated with a separate medical VQA dataset.

ConanVarvar: a versatile tool for the detection of large syndromic copy number variation from whole-genome sequencing data.

BACKGROUND: A wide range of tools are available for the detection of copy number variants (CNVs) from whole-genome sequencing (WGS) data. However, none of them focus on clinically-relevant CNVs, such as those that are associated with known genetic syndromes. Such variants are often large in size, typically 1-5 Mb, but currently available CNV callers have been developed and benchmarked for the discovery of smaller variants. Thus, the ability of these programs to detect tens of real syndromic CNVs remains largely unknown. RESULTS: Here we present ConanVarvar, a tool which implements a complete workflow for the targeted analysis of large germline CNVs from WGS data. ConanVarvar comes with an intuitive R Shiny graphical user interface and annotates identified variants with information about 56 associated syndromic conditions. We benchmarked ConanVarvar and four other programs on a dataset containing real and simulated syndromic CNVs larger than 1 Mb. In comparison to other tools, ConanVarvar reports 10-30 times less false-positive variants without compromising sensitivity and is quicker to run, especially on large batches of samples. CONCLUSIONS: ConanVarvar is a useful instrument for primary analysis in disease sequencing studies, where large CNVs could be the cause of disease.

Vision-Language Transformer for Interpretable Pathology Visual Question Answering.

Pathology visual question answering (PathVQA) attempts to answer a medical question posed by pathology images. Despite its great potential in healthcare, it is not widely adopted because it requires interactions on both the image (vision) and question (language) to generate an answer. Existing methods focused on treating vision and language features independently, which were unable to capture the high and low-level interactions that are required for VQA. Further, these methods failed to offer capabilities to interpret the retrieved answers, which are obscure to humans where the models' interpretability to justify the retrieved answers has remained largely unexplored. Motivated by these limitations, we introduce a vision-language transformer that embeds vision (images) and language (questions) features for an interpretable PathVQA. We present an interpretable transformer-based Path-VQA (TraP-VQA), where we embed transformers' encoder layers with vision and language features extracted using pre-trained CNN and domain-specific language model (LM), respectively. A decoder layer is then embedded to upsample the encoded features for the final prediction for PathVQA. Our experiments showed that our TraP-VQA outperformed the state-of-the-art comparative methods with public PathVQA dataset. Our experiments validated the robustness of our model on another medical VQA dataset, and the ablation study demonstrated the capability of our integrated transformer-based vision-language model for PathVQA. Finally, we present the visualization results of both text and images, which explain the reason for a retrieved answer in PathVQA.

Benchmarking for biomedical natural language processing tasks with a domain specific ALBERT.

BACKGROUND: The abundance of biomedical text data coupled with advances in natural language processing (NLP) is resulting in novel biomedical NLP (BioNLP) applications. These NLP applications, or tasks, are reliant on the availability of domain-specific language models (LMs) that are trained on a massive amount of data. Most of the existing domain-specific LMs adopted bidirectional encoder representations from transformers (BERT) architecture which has limitations, and their generalizability is unproven as there is an absence of baseline results among common BioNLP tasks. RESULTS: We present 8 variants of BioALBERT, a domain-specific adaptation of a lite bidirectional encoder representations from transformers (ALBERT), trained on biomedical (PubMed and PubMed Central) and clinical (MIMIC-III) corpora and fine-tuned for 6 different tasks across 20 benchmark datasets. Experiments show that a large variant of BioALBERT trained on PubMed outperforms the state-of-the-art on named-entity recognition (+ 11.09% BLURB score improvement), relation extraction (+ 0.80% BLURB score), sentence similarity (+ 1.05% BLURB score), document classification (+ 0.62% F1-score), and question answering (+ 2.83% BLURB score). It represents a new state-of-the-art in 5 out of 6 benchmark BioNLP tasks. CONCLUSIONS: The large variant of BioALBERT trained on PubMed achieved a higher BLURB score than previous state-of-the-art models on 5 of the 6 benchmark BioNLP tasks. Depending on the task, 5 different variants of BioALBERT outperformed previous state-of-the-art models on 17 of the 20 benchmark datasets, showing that our model is robust and generalizable in the common BioNLP tasks. We have made BioALBERT freely available which will help the BioNLP community avoid computational cost of training and establish a new set of baselines for future efforts across a broad range of BioNLP tasks.

Deep convolutional neural networks based ECG beats classification to diagnose cardiovascular conditions.

Medical practitioners need to understand the critical features of ECG beats to diagnose and identify cardiovascular conditions accurately. This would be greatly facilitated by identifying the significant features of frequency components in temporal ECG wave-forms using computational methods. In this study, we have proposed a novel ECG beat classifier based on a customized VGG16-based Convolution Neural Network (CNN) that uses the time-frequency representation of temporal ECG, and a method to identify the contribution of interpretable ECG frequencies when classifying based on the SHapley Additive exPlanations (SHAP) values. We applied our model to the MIT-BIH arrhythmia dataset to classify the ECG beats and to characterise of the beats frequencies. This model was evaluated with two advanced time-frequency analysis methods. Our results indicated that for 2-4 classes our proposed model achieves a classification accuracy of 100% and for 5 classes it achieves a classification accuracy of 99.90%. We have also tested the proposed model using premature ventricular contraction beats from the American Heart Association (AHA) database and normal beats from Lobachevsky University Electrocardiography database (LUDB) and obtained a classification accuracy of 99.91% for the 5-classes case. In addition, SHAP value increased the interpretability of the ECG frequency features. Thus, this model could be applicable to the automation of the cardiovascular diagnosis system and could be used by clinicians.

How did socio-demographic status and personal attributes influence compliance to COVID-19 preventive behaviours during the early outbreak in Japan? Lessons for pandemic management.

This study focuses on how socio-demographic status and personal attributes influence self-protective behaviours during a pandemic, with protection behaviours being assessed through three perspectives - social distancing, personal protection behaviour and social responsibility awareness. The research considers a publicly available and recently collected dataset on Japanese citizens during the COVID-19 early outbreak and utilises a data analysis framework combining Classification and Regression Tree (CART), a data mining approach, and regression analysis to gain deep insights. The analysis reveals Socio-demographic attributes - sex, marital family status and having children - as having played an influential role in Japanese citizens' abiding by the COVID-19 protection behaviours. Especially women with children are noted as more conscious than their male counterparts. Work status also appears to have some impact concerning social distancing. Trust in government also appears as a significant factor. The analysis further identifies smoking behaviour as a factor characterising subjective prevention actions with non-smokers or less-frequent smokers being more compliant to the protection behaviours. Overall, the findings imply the need of public policy campaigning to account for variations in protection behaviour due to socio-demographic and personal attributes during pandemics and national emergencies.

COVIDSenti: A Large-Scale Benchmark Twitter Data Set for COVID-19 Sentiment Analysis.

Social media (and the world at large) have been awash with news of the COVID-19 pandemic. With the passage of time, news and awareness about COVID-19 spread like the pandemic itself, with an explosion of messages, updates, videos, and posts. Mass hysteria manifest as another concern in addition to the health risk that COVID-19 presented. Predictably, public panic soon followed, mostly due to misconceptions, a lack of information, or sometimes outright misinformation about COVID-19 and its impacts. It is thus timely and important to conduct an ex post facto assessment of the early information flows during the pandemic on social media, as well as a case study of evolving public opinion on social media which is of general interest. This study aims to inform policy that can be applied to social media platforms; for example, determining what degree of moderation is necessary to curtail misinformation on social media. This study also analyzes views concerning COVID-19 by focusing on people who interact and share social media on Twitter. As a platform for our experiments, we present a new large-scale sentiment data set COVIDSENTI, which consists of 90 000 COVID-19-related tweets collected in the early stages of the pandemic, from February to March 2020. The tweets have been labeled into positive, negative, and neutral sentiment classes. We analyzed the collected tweets for sentiment classification using different sets of features and classifiers. Negative opinion played an important role in conditioning public sentiment, for instance, we observed that people favored lockdown earlier in the pandemic; however, as expected, sentiment shifted by mid-March. Our study supports the view that there is a need to develop a proactive and agile public health presence to combat the spread of negative sentiment on social media following a pandemic.

Delayed recruiting of TPD52 to lipid droplets - evidence for a "second wave" of lipid droplet-associated proteins that respond to altered lipid storage induced by Brefeldin A treatment.

Tumor protein D52 (TPD52) is amplified and overexpressed in breast and prostate cancers which are frequently characterised by dysregulated lipid storage and metabolism. TPD52 expression increases lipid storage in mouse 3T3 fibroblasts, and co-distributes with the Golgi marker GM130 and lipid droplets (LDs). We examined the effects of Brefeldin A (BFA), a fungal metabolite known to disrupt the Golgi structure, in TPD52-expressing 3T3 cells, and in human AU565 and HMC-1-8 breast cancer cells that endogenously express TPD52. Five-hour BFA treatment reduced median LD numbers, but increased LD sizes. TPD52 knockdown decreased both LD sizes and numbers, and blunted BFA's effects on LD numbers. Following BFA treatment for 1-3 hours, TPD52 co-localised with the trans-Golgi network protein syntaxin 6, but after 5 hours BFA treatment, TPD52 showed increased co-localisation with LDs, which was disrupted by microtubule depolymerising agent nocodazole. BFA treatment also increased perilipin (PLIN) family protein PLIN3 but reduced PLIN2 detection at LDs in TPD52-expressing 3T3 cells, with PLIN3 recruitment to LDs preceding that of TPD52. An N-terminally deleted HA-TPD52 mutant (residues 40-184) almost exclusively targeted to LDs in both vehicle and BFA treated cells. In summary, delayed recruitment of TPD52 to LDs suggests that TPD52 participates in a temporal hierarchy of LD-associated proteins that responds to altered LD packaging requirements induced by BFA treatment.

Promotion of ß-catenin/Foxo1 signaling ameliorates renal interstitial fibrosis.

Transforming growth factor ß (TGF-ß) is the key cytokine involved in causing fibrosis through cross-talk with major profibrotic pathways. However, inhibition of TGF-ß to prevent fibrosis would also abrogate its anti-inflammatory and wound-healing effects. ß-catenin is a common co-factor in most TGF-ß signaling pathways. ß-catenin binds to T-cell factor (TCF) to activate profibrotic genes and binds to Forkhead box O (Foxo) to promote cell survival under oxidative stress. Using a proximity ligation assay in human kidney biopsies, we found that ß-catenin/Foxo interactions were higher in kidney with little fibrosis, whereas ß-catenin/TCF interactions were upregulated in the kidney of patients with fibrosis. We hypothesised that ß-catenin/Foxo is protective against kidney fibrosis. We found that Foxo1 protected against rhTGF-ß1-induced profibrotic protein expression using a CRISPR/cas9 knockout of Foxo1 or TCF1 in murine kidney tubular epithelial C1.1 cells. Co-administration of TGF-ß with a small molecule inhibitor of ß-catenin/TCF (ICG-001), protected against kidney fibrosis in unilateral ureteral obstruction. Collectively, our human, animal and in vitro findings suggest ß-catenin/Foxo as a therapeutic target in kidney fibrosis.

MatCol: a tool to measure fluorescence signal colocalisation in biological systems.

Protein colocalisation is often studied using pixel intensity-based coefficients such as Pearson, Manders, Li or Costes. However, these methods cannot be used to study object-based colocalisations in biological systems. Therefore, a novel method is required to automatically identify regions of fluorescent signal in two channels, identify the co-located parts of these regions, and calculate the statistical significance of the colocalisation. We have developed MatCol to address these needs. MatCol can be used to visualise protein and/or DNA colocalisations and fine tune user-defined parameters for the colocalisation analysis, including the application of median or Wiener filtering to improve the signal to noise ratio. Command-line execution allows batch processing of multiple images. Users can also calculate the statistical significance of the observed object colocalisations compared to overlap by random chance using Student's t-test. We validated MatCol in a biological setting. The colocalisations of telomeric DNA and TRF2 protein or TRF2 and PML proteins in >350 nuclei derived from three different cell lines revealed a highly significant correlation between manual and MatCol identification of colocalisations (linear regression R2 = 0.81, P < 0.0001). MatCol has the ability to replace manual colocalisation counting, and the potential to be applied to a wide range of biological areas.

Automated classification and characterization of the mitotic spindle following knockdown of a mitosis-related protein.

BACKGROUND: Cell division (mitosis) results in the equal segregation of chromosomes between two daughter cells. The mitotic spindle plays a pivotal role in chromosome alignment and segregation during metaphase and anaphase. Structural or functional errors of this spindle can cause aneuploidy, a hallmark of many cancers. To investigate if a given protein associates with the mitotic spindle and regulates its assembly, stability, or function, fluorescence microscopy can be performed to determine if disruption of that protein induces phenotypes indicative of spindle dysfunction. Importantly, functional disruption of proteins with specific roles during mitosis can lead to cancer cell death by inducing mitotic insult. However, there is a lack of automated computational tools to detect and quantify the effects of such disruption on spindle integrity. RESULTS: We developed the image analysis software tool MatQuantify, which detects both large-scale and subtle structural changes in the spindle or DNA and can be used to statistically compare the effects of different treatments. MatQuantify can quantify various physical properties extracted from fluorescence microscopy images, such as area, lengths of various components, perimeter, eccentricity, fractal dimension, satellite objects and orientation. It can also measure textual properties including entropy, intensities and the standard deviation of intensities. Using MatQuantify, we studied the effect of knocking down the protein clathrin heavy chain (CHC) on the mitotic spindle. We analysed 217 microscopy images of untreated metaphase cells, 172 images of metaphase cells transfected with small interfering RNAs targeting the luciferase gene (as a negative control), and 230 images of metaphase cells depleted of CHC. Using the quantified data, we trained 23 supervised machine learning classification algorithms. The Support Vector Machine learning algorithm was the most accurate method (accuracy: 85.1%; area under the curve: 0.92) for classifying a spindle image. The Kruskal-Wallis and Tukey-Kramer tests demonstrated that solidity, compactness, eccentricity, extent, mean intensity and number of satellite objects (multipolar spindles) significantly differed between CHC-depleted cells and untreated/luciferase-knockdown cells. CONCLUSION: MatQuantify enables automated quantitative analysis of images of mitotic spindles. Using this tool, researchers can unambiguously test if disruption of a protein-of-interest changes metaphase spindle maintenance and thereby affects mitosis.

Benchmarking database performance for genomic data.

Genomic regions represent features such as gene annotations, transcription factor binding sites and epigenetic modifications. Performing various genomic operations such as identifying overlapping/non-overlapping regions or nearest gene annotations are common research needs. The data can be saved in a database system for easy management, however, there is no comprehensive database built-in algorithm at present to identify overlapping regions. Therefore I have developed a novel region-mapping (RegMap) SQL-based algorithm to perform genomic operations and have benchmarked the performance of different databases. Benchmarking identified that PostgreSQL extracts overlapping regions much faster than MySQL. Insertion and data uploads in PostgreSQL were also better, although general searching capability of both databases was almost equivalent. In addition, using the algorithm pair-wise, overlaps of >1000 datasets of transcription factor binding sites and histone marks, collected from previous publications, were reported and it was found that HNF4G significantly co-locates with cohesin subunit STAG1 (SA1).Inc.

Bioinformatic analysis of cis-regulatory interactions between progesterone and estrogen receptors in breast cancer.

Chromatin factors interact with each other in a cell and sequence-specific manner in order to regulate transcription and a wealth of publically available datasets exists describing the genomic locations of these interactions. Our recently published BiSA (Binding Sites Analyser) database contains transcription factor binding locations and epigenetic modifications collected from published studies and provides tools to analyse stored and imported data. Using BiSA we investigated the overlapping cis-regulatory role of estrogen receptor alpha (ERα) and progesterone receptor (PR) in the T-47D breast cancer cell line. We found that ERα binding sites overlap with a subset of PR binding sites. To investigate further, we re-analysed raw data to remove any biases introduced by the use of distinct tools in the original publications. We identified 22,152 PR and 18,560 ERα binding sites (<5% false discovery rate) with 4,358 overlapping regions among the two datasets. BiSA statistical analysis revealed a non-significant overall overlap correlation between the two factors, suggesting that ERα and PR are not partner factors and do not require each other for binding to occur. However, Monte Carlo simulation by Binary Interval Search (BITS), Relevant Distance, Absolute Distance, Jaccard and Projection tests by Genometricorr revealed a statistically significant spatial correlation of binding regions on chromosome between the two factors. Motif analysis revealed that the shared binding regions were enriched with binding motifs for ERα, PR and a number of other transcription and pioneer factors. Some of these factors are known to co-locate with ERα and PR binding. Therefore spatially close proximity of ERα binding sites with PR binding sites suggests that ERα and PR, in general function independently at the molecular level, but that their activities converge on a specific subset of transcriptional targets.

Binding sites analyser (BiSA): software for genomic binding sites archiving and overlap analysis.

Genome-wide mapping of transcription factor binding and histone modification reveals complex patterns of interactions. Identifying overlaps in binding patterns by different factors is a major objective of genomic studies, but existing methods to archive large numbers of datasets in a personalised database lack sophistication and utility. Therefore we have developed transcription factor DNA binding site analyser software (BiSA), for archiving of binding regions and easy identification of overlap with or proximity to other regions of interest. Analysis results can be restricted by chromosome or base pair overlap between regions or maximum distance between binding peaks. BiSA is capable of reporting overlapping regions that share common base pairs; regions that are nearby; regions that are not overlapping; and average region sizes. BiSA can identify genes located near binding regions of interest, genomic features near a gene or locus of interest and statistical significance of overlapping regions can also be reported. Overlapping results can be visualized as Venn diagrams. A major strength of BiSA is that it is supported by a comprehensive database of publicly available transcription factor binding sites and histone modifications, which can be directly compared to user data. The documentation and source code are available on http://bisa.sourceforge.net.

Open source tools for management and archiving of digital microscopy data to allow integration with patient pathology and treatment information.

BACKGROUND: Virtual microscopy includes digitisation of histology slides and the use of computer technologies for complex investigation of diseases such as cancer. However, automated image analysis, or website publishing of such digital images, is hampered by their large file sizes. RESULTS: We have developed two Java based open source tools: Snapshot Creator and NDPI-Splitter. Snapshot Creator converts a portion of a large digital slide into a desired quality JPEG image. The image is linked to the patient's clinical and treatment information in a customised open source cancer data management software (Caisis) in use at the Australian Breast Cancer Tissue Bank (ABCTB) and then published on the ABCTB website (http://www.abctb.org.au) using Deep Zoom open source technology. Using the ABCTB online search engine, digital images can be searched by defining various criteria such as cancer type, or biomarkers expressed. NDPI-Splitter splits a large image file into smaller sections of TIFF images so that they can be easily analysed by image analysis software such as Metamorph or Matlab. NDPI-Splitter also has the capacity to filter out empty images. CONCLUSIONS: Snapshot Creator and NDPI-Splitter are novel open source Java tools. They convert digital slides into files of smaller size for further processing. In conjunction with other open source tools such as Deep Zoom and Caisis, this suite of tools is used for the management and archiving of digital microscopy images, enabling digitised images to be explored and zoomed online. Our online image repository also has the capacity to be used as a teaching resource. These tools also enable large files to be sectioned for image analysis. VIRTUAL SLIDES: The virtual slide(s) for this article can be found here: http://www.diagnosticpathology.diagnomx.eu/vs/5330903258483934.

Development of a data entry auditing protocol and quality assurance for a tissue bank database.

Human transcription error is an acknowledged risk when extracting information from paper records for entry into a database. For a tissue bank, it is critical that accurate data are provided to researchers with approved access to tissue bank material. The challenges of tissue bank data collection include manual extraction of data from complex medical reports that are accessed from a number of sources and that differ in style and layout. As a quality assurance measure, the Breast Cancer Tissue Bank (http:\\www.abctb.org.au) has implemented an auditing protocol and in order to efficiently execute the process, has developed an open source database plug-in tool (eAuditor) to assist in auditing of data held in our tissue bank database. Using eAuditor, we have identified that human entry errors range from 0.01% when entering donor's clinical follow-up details, to 0.53% when entering pathological details, highlighting the importance of an audit protocol tool such as eAuditor in a tissue bank database. eAuditor was developed and tested on the Caisis open source clinical-research database; however, it can be integrated in other databases where similar functionality is required.

Electronic biorepository application system: web-based software to manage receipt, peer review, and approval of researcher applications to a biobank.

The importance of suitably characterized and preserved biospecimens for research is acknowledged, yet providing information about the availability of biospecimens and associated data, responding to enquiries from researchers, processing applications for material, and seeking independent scientific review of proposed projects are complex and time-consuming processes. Most biorepositories operate as not-for-profit entities; therefore, cost containment is a major consideration. We identified that online systematizing and automation of all of the tasks associated with application management could reduce the administrative workload and therefore reduce costs and improve the efficiency of a biobank. Accordingly, we have developed a Web-based electronic Biorepository Application System (eBAS) that allows researchers to search for suitable material from the biobank database, submit an online expression of interest, and complete all the information required for a full application. Peer review is also managed through eBAS. Implementation of eBAS has streamlined application management and external peer review of researcher applications, and has facilitated automated record storage and management. This approach has potential to reduce the costs and complexities of administering researcher applications. We have also linked eBAS to an open-source clinical research and specimen management database, Caisis.