RESUMO
Understanding causality is a longstanding goal across many different domains. Different articles, such as those published in medical journals, disseminate newly discovered knowledge that is often causal. In this paper, we use this intuition to build a model that leverages causal relations to unearth factors related to Sjögren's syndrome from biomedical literature. Sjögren's syndrome is an autoimmune disease affecting up to 3.1 million Americans. Due to the uncommon nature of the illness, symptoms across different specialties coupled with common symptoms of other autoimmune conditions such as rheumatoid arthritis, it is difficult for clinicians to diagnose the disease timely. Due to the lack of a dedicated dataset for causal relationships built from biomedical literature, we propose a transfer learning-based approach, where the relationship extraction model is trained on a wide variety of datasets. We conduct an empirical analysis of numerous neural network architectures and data transfer strategies for causal relation extraction. By conducting experiments with various contextual embedding layers and architectural components, we show that an ELECTRA-based sentence-level relation extraction model generalizes better than other architectures across varying web-based sources and annotation strategies. We use this empirical observation to create a pipeline for identifying causal sentences from literature text, extracting the causal relationships from causal sentences, and building a causal network consisting of latent factors related to Sjögren's syndrome. We show that our approach can retrieve such factors with high precision and recall values. Comparative experiments show that this approach leads to 25% improvement in retrieval F1-score compared to several state-of-the-art biomedical models, including BioBERT and Gram-CNN. We apply this model to a corpus of research articles related to Sjögren's syndrome collected from PubMed to create a causal network for Sjögren's syndrome. The proposed causal network for Sjögren's syndrome will potentially help clinicians with a holistic knowledge base for faster diagnosis.
RESUMO
Purpose: Prior studies show convolutional neural networks predicting self-reported race using x-rays of chest, hand and spine, chest computed tomography, and mammogram. We seek an understanding of the mechanism that reveals race within x-ray images, investigating the possibility that race is not predicted using the physical structure in x-ray images but is embedded in the grayscale pixel intensities. Approach: Retrospective full year 2021, 298,827 AP/PA chest x-ray images from 3 academic health centers across the United States and MIMIC-CXR, labeled by self-reported race, were used in this study. The image structure is removed by summing the number of each grayscale value and scaling to percent per image (PPI). The resulting data are tested using multivariate analysis of variance (MANOVA) with Bonferroni multiple-comparison adjustment and class-balanced MANOVA. Machine learning (ML) feed-forward networks (FFN) and decision trees were built to predict race (binary Black or White and binary Black or other) using only grayscale value counts. Stratified analysis by body mass index, age, sex, gender, patient type, make/model of scanner, exposure, and kilovoltage peak setting was run to study the impact of these factors on race prediction following the same methodology. Results: MANOVA rejects the null hypothesis that classes are the same with 95% confidence (F 7.38, P<0.0001) and balanced MANOVA (F 2.02, P<0.0001). The best FFN performance is limited [area under the receiver operating characteristic (AUROC) of 69.18%]. Gradient boosted trees predict self-reported race using grayscale PPI (AUROC 77.24%). Conclusions: Within chest x-rays, pixel intensity value counts alone are statistically significant indicators and enough for ML classification tasks of patient self-reported race.
RESUMO
BACKGROUND: Understanding the genome of Cannabis sativa holds significant scientific value due to the multi-faceted therapeutic nature of the plant. Links from cannabis gene to therapeutic property are important to establish gene targets for the optimization of specific therapeutic properties through selective breeding of cannabis strains. Our work establishes a resource for quickly obtaining a complete set of therapeutic properties and genes associated with any known cannabis chemical constituent, as well as relevant literature. METHODS: State-of-the-art natural language processing (NLP) was used to automatically extract information from many cannabis-related publications, thus producing an undirected multipartite weighted-edge paragraph co-occurrence relationship network composed of two relationship types, gene-chemical and chemical property. We also developed an interactive application to visualize sub-graphs of manageable size. RESULTS: Two hundred thirty-four cannabis constituent chemicals, 352 therapeutic properties, and 124 genes from the Cannabis sativa genome form a multipartite network graph which transforms 29,817 cannabis-related research documents from PubMed Central into an easy to visualize and explore network format. CONCLUSION: Use of our network replaces time-consuming and labor intensive manual extraction of information from the large amount of available cannabis literature. This streamlined information retrieval process will enhance the activities of cannabis breeders, cannabis researchers, organic biochemists, pharmaceutical researchers and scientists in many other disciplines.
