Data Extraction Pipeline Research Articles

Versatile Deep Learning Pipeline for Transferable Chemical Data Extraction.

Chemical information disseminated in scientific documents offers an untapped potential for deep learning-assisted insights and breakthroughs. Automated extraction efforts have shifted from resource-intensive manual extraction toward applying machine learning methods to streamline chemical data extraction. While current extraction models and pipelines have ushered in notable efficiency improvements, they often exhibit modest performance, compromising the accuracy of predictive models trained on extracted data. Further, current chemical pipelines lack both transferability─where a model trained on one task can be adapted to another relevant task with limited examples─and extensibility, which enables seamless adaptability for new extraction tasks. Addressing these gaps, we present ChemREL, a versatile chemical data extraction pipeline emphasizing performance, transferability, and extensibility. ChemREL utilizes a custom, diverse data set of chemical documents, labeled through an active learning strategy to extract two properties: normal melting point and lethal dose 50 (LD50). The normal melting point is selected for its prevalence in diverse contexts and wider literature, serving as the foundation for pipeline training. In contrast, LD50 evaluates the pipeline's transferability to an unrelated property, underscoring variance in its biological nature, toxicological context, and units, among other differences. With pretraining and fine-tuning, our pipeline outperforms existing methods and GPT-4, achieving F1-scores of 96.1% for entity identification and 97.0% for relation mapping, culminating in an overall F1-score of 95.4%. More importantly, ChemREL displays high transferability, effectively transitioning from melting point extraction to LD50 extraction with 10 randomly selected training documents. Released as an open-source package, ChemREL aims to broaden access to chemical data extraction, enabling the construction of expansive relational data sets that propel discovery.

A general-purpose material property data extraction pipeline from large polymer corpora using natural language processing

The ever-increasing number of materials science articles makes it hard to infer chemistry-structure-property relations from literature. We used natural language processing methods to automatically extract material property data from the abstracts of polymer literature. As a component of our pipeline, we trained MaterialsBERT, a language model, using 2.4 million materials science abstracts, which outperforms other baseline models in three out of five named entity recognition datasets. Using this pipeline, we obtained ~300,000 material property records from ~130,000 abstracts in 60 hours. The extracted data was analyzed for a diverse range of applications such as fuel cells, supercapacitors, and polymer solar cells to recover non-trivial insights. The data extracted through our pipeline is made available at polymerscholar.org which can be used to locate material property data recorded in abstracts. This work demonstrates the feasibility of an automatic pipeline that starts from published literature and ends with extracted material property information.

Parselmouth for bioacoustics: Analysis pipelines for seal vocalizations

Every empirical research project includes bottlenecks at various levels. In bioacoustics, one of these time-consuming bottlenecks corresponds to the step of transforming a long stream of audio into acoustic properties of specific sounds. Here, we describe a data-extraction pipeline which integrates manual annotation with Parselmouth’s powerful computational analyses. This semi-supervised method allows extracting a large volume of sound features with limited repetitive human “point and click.” We illustrate this using recently published empirical research, where we focused on vocal production learning and plasticity in pinnipeds. Faced with a species capable of imitating sounds, fully automatic methods may misclassify individuals (because of imitation), while the large number of calls make fully manual approaches suboptimal and error-prone. Focusing on early vocal development, we tested 1–3 weeks-old harbor seal pups (Phoca vitulina). Noise playbacks served to induce seal pups to shift their fundamental frequency. Pups' spontaneous calls were recorded while exposed to bandpass-filtered noise, which spanned and masked the animals’ fundamental frequency range. After a summary manual annotation of calls’ boundaries, Parselmouth identified these boundaries in the files, and automatically extracted multiple sound parameters. Based on this, we found that pups modified their vocalizations by lowering their fundamental frequency in response to noise.

BatteryDataExtractor: battery-aware text-mining software embedded with BERT models.

Due to the massive growth of scientific publications, literature mining is becoming increasingly popular for researchers to thoroughly explore scientific text and extract such data to create new databases or augment existing databases. Efforts in literature-mining software design and implementation have improved text-mining productivity, but most of the toolkits that mine text are based on traditional machine-learning-algorithms which hinder the performance of downstream text-mining tasks. Natural-language processing (NLP) and text-mining technologies have seen a rapid development since the release of transformer models, such as bidirectional encoder representations from transformers (BERT). Upgrading rule-based or machine-learning-based literature-mining toolkits by embedding transformer models into the software is therefore likely to improve their text-mining performance. To this end, we release a Python-based literature-mining toolkit for the field of battery materials, BatteryDataExtractor, which involves the embedding of BatteryBERT models in its automated data-extraction pipeline. This pipeline employs BERT models for token-classification tasks, such as abbreviation detection, part-of-speech tagging, and chemical-named-entity recognition, as well as new double-turn question-answering data-extraction models for auto-generating repositories of inter-related material and property data as well as general information. We demonstrate that BatteryDataExtractor exhibits state-of-the-art performance on the evaluation data sets for both token classification and automated data extraction. To aid the use of BatteryDataExtractor, its code is provided as open-source software, with associated documentation to serve as a user guide.

Application of Cognitive Automation to Structuring Data, Driving Existing Business Models, and Creating Value between Legacy Industries

To build high quality datasets and unlock the value of unstructured data, a systematic approach for data capture is necessary. Cognitive automation (CA), that is, automation of processes with artificial intelligence (AI), enables the information extraction from unstructured data to provide relevant insights and further processing with AI. This study provides an overview of this new technology and shows how it can be used to transform existing business models. Our case studies in the insurance auditing, healthcare, and banking industries show the potential managerial impact of CA, which prepares these legacy industries for their digital future’s challenges and opportunities. We present the novel data extraction pipeline for textual and visual data and demonstrate its efficiency in extracting information from the company’s unstructured data. We show its performance in quality, cost, and time compared with current industry standards and provide management insights for business applications using CA.

Abstract 13948: Deep Learning Identification of Brain Structural Atrophy Associated With Heart Failure With Preserved Ejection Fraction Among Patients With Preexisting Dementia Using Clinical Imaging

Background: Heart failure with preserved ejection fraction (HFpEF) and dementia share several cardiovascular risk factors and have been independently associated with brain atrophy. However, it is unclear how HFpEF affects brain morphology in patients with preexisting dementia. Advances in bioinformatics data extraction pipelines, quantitative neuroimaging, and deep learning, have made possible the analysis of the effects of cardiovascular disease on clinical neuroimaging in large datasets. Hypothesis: We hypothesized that HFpEF in the setting of dementia would be associated with distinct brain morphology imaging changes compared to dementia in the absence of HFpEF. Methods: We used the bioinformatics tools at Vanderbilt University Medical Center to extract and analyze deidentified clinical neuroimaging from patients with a diagnosis of dementia with and without coexisting HFpEF from the electronic health record. We identified a total of 30 patients with HFpEF and dementia who had high-resolution, clinical neuroimaging as well as 301 age- and sex-matched controls. We used an automated pipeline to segment the brain into 132 distinct brain regions and estimate their volume to find structural differences associated with HFpEF. Results: We found six regions with significant atrophy associated with HFpEF: accumbens area, amygdala, posterior insula, anterior orbital gyrus, angular gyrus, and cerebellar white matter. There were no regions with atrophy inversely associated with HFpEF (Figure 1). Conclusions: Here, we show that patients with HFpEF have a distinct neuroimaging signature on patients with preexisting dementia. A possible mechanism of injury for this region and the subsequent atrophy has been linked to high aortic stiffening secondary to chronic cardiovascular disease and its risk factors, resulting in end-organ damage due to barotrauma and shear forces.

Adaptive and Background-Aware GAL4 Expression Enhancement of Co-registered Confocal Microscopy Images.

GAL4 gene expression imaging using confocal microscopy is a common and powerful technique used to study the nervous system of a model organism such as Drosophila melanogaster. Recent research projects focused on high throughput screenings of thousands of different driver lines, resulting in large image databases. The amount of data generated makes manual assessment tedious or even impossible. The first and most important step in any automatic image processing and data extraction pipeline is to enhance areas with relevant signal. However, data acquired via high throughput imaging tends to be less then ideal for this task, often showing high amounts of background signal. Furthermore, neuronal structures and in particular thin and elongated projections with a weak staining signal are easily lost. In this paper we present a method for enhancing the relevant signal by utilizing a Hessian-based filter to augment thin and weak tube-like structures in the image. To get optimal results, we present a novel adaptive background-aware enhancement filter parametrized with the local background intensity, which is estimated based on a common background model. We also integrate recent research on adaptive image enhancement into our approach, allowing us to propose an effective solution for known problems present in confocal microscopy images. We provide an evaluation based on annotated image data and compare our results against current state-of-the-art algorithms. The results show that our algorithm clearly outperforms the existing solutions.

Argonaut

Crowdsourced workflows are used in research and industry to solve a variety of tasks. The databases community has used crowd workers in query operators/optimization and for tasks such as entity resolution. Such research utilizes microtasks where crowd workers are asked to answer simple yes/no or multiple choice questions with little training. Typically, microtasks are used with voting algorithms to combine redundant responses from multiple crowd workers to achieve result quality. Microtasks are powerful, but fail in cases where larger context (e.g., domain knowledge) or significant time investment is needed to solve a problem, for example in large-document structured data extraction. In this paper, we consider context-heavy data processing tasks that may require many hours of work, and refer to such tasks as macrotasks. Leveraging the infrastructure and worker pools of existing crowdsourcing platforms, we automate macrotask scheduling, evaluation, and pay scales. A key challenge in macrotask-powered work, however, is evaluating the quality of a worker's output, since ground truth is seldom available and redundancy-based quality control schemes are impractical. We present Argonaut, a framework that improves macrotask powered work quality using a hierarchical review. Argonaut uses a predictive model of worker quality to select trusted workers to perform review, and a separate predictive model of task quality to decide which tasks to review. Finally, Argonaut can identify the ideal trade-off between a single phase of review and multiple phases of review given a constrained review budget in order to maximize overall output quality. We evaluate an industrial use of Argonaut to power a structured data extraction pipeline that has utilized over half a million hours of crowd worker input to complete millions of macrotasks. We show that Argonaut can capture up to 118% more errors than random spot-check reviews in review budget-constrained environments with up to two review layers.

Abstract 680: What Matters Most, Statin Intensity or Achieved LDL? - Evaluating Concordance of AHA/ACC Guidelines for Statin Use with Practice Outcomes at Stanford Hospital & Clinics

Introduction: Recommendations from the 2013 American College of Cardiology/American Heart Association Guidelines on the Assessment of Cardiovascular Risk have had a mixed reception. The goal of our study was to evaluate practice-based evidence from Stanford Hospital using structured and unstructured bioinformatics data to determine whether achieved LDL or statin intensity was more predictive of major adverse cardiac events (MACE). Methods: We performed a retrospective cohort analysis of all adult patients >= 21 years of age prescribed statin therapy between 1996 and 2015. Demographic and clinical variables were extracted from coded data and unstructured clinical text using a validated data extraction pipeline. To account for possible treatment selection bias we performed 1:1 propensity score matching to produce a patient cohort with similar demographic, comorbid disease, and laboratory value distributions between treatment groups. Mcnemar’s Chi-squared and conditional logistic regression analyses were performed to identify variables predictive of MACE. Results: We identified 7,378 adults on statin therapy with complete data. Patients were followed on average for 3.3 years, 56% percent were male, 55% were Caucasian, and mean age was 64 years. Patients had a high burden of comorbidities (85% with HTN, 64% with Type II Diabetes, 58% with CAD, 42% with CHF, and 32% with PAD). Patients on high-intensity statins were matched with those on low-/moderate-intensity statins, producing a cohort of 2,056 patients with equal distributions of comorbidities, co-prescriptions and LDL measurements. In this well-balanced cohort, statin intensity did not predict MACE (P=0.29) while achieved LDL was a significant predictor of outcomes (P=0.006). OR for MACE was 0.73 (95% CI 0.65-0.82) in patients with LDL of <=70 versus those with LDL >100. OR for MACE for LDL <=70 versus LDL >= 130 was 0.32 (95% CI 0.28-0.38). Conclusions: Using a retrospective matched cohort design we found that achieved LDL was a significant predictor of MACE outcome, while statin intensity was not. This finding implies that despite recent guidelines, statin therapy should continue to be aimed at achieving specific LDL targets rather than arbitrarily selecting statins based on intensity alone.

Building the Coleoptera tree‐of‐life for >8000 species: composition of public DNA data and fit with Linnaean classification

AbstractThe species representation of public databases is growing rapidly and permits increasingly detailed phylogenetic inferences. We present a supermatrix based on all gene sequences of Coleoptera available in Genbank for two nuclear (18S and 28S rRNA) and two mitochondrial (rrnL and cox1) genes. After filtering for unique species names and the addition of ˜2000 unpublished sequences for cox1 and 18S rRNA, the resulting data matrix included 8441 species‐level terminals and 6600 aligned nucleotide positions. The concatenated matrix represents the equivalent of 2.17% of the 390 000 described species of Coleoptera and includes 152 beetle families. The remaining 29 families constitute small lineages with ˜250 known species in total. Taxonomic coverage remains low for several major lineages, including Buprestidae (0.16% of described species), Staphylinidae (1.03%), Tenebrionidae (0.90%) and Cerambycidae (0.58%). The current taxon sampling was strongly biased towards the Northern Hemisphere. Phylogenetic trees obtained from the supermatrix were in very good agreement with the Linnaean classification, in particular at the family level, but lower for the subfamily and lowest for the genus level. The topology supports the basal split of Derodontidae and Scirtoidea from the remaining Polyphaga, and the broad paraphyly of Cucujoidea. The data extraction pipeline and detailed tree provide a framework for placement of any new sequences, including environmental samples, into a DNA‐based classification system of Coleoptera.

An eUtils toolset and its use for creating a pipeline to link genomics and proteomics analyses to domain-specific biomedical literature

BackgroundNumerous biomedical software applications access databases maintained by the US National Center for Biotechnology Information (NCBI). To ease software automation, NCBI provides a powerful but complex Web-service-based programming interface, eUtils. This paper describes a toolset that simplifies eUtils use through a graphical front-end that can be used by non-programmers to construct data-extraction pipelines. The front-end relies on a code library that provides high-level wrappers around eUtils functions, and which is distributed as open-source, allowing customization and enhancement by individuals with programming skills.MethodsWe initially created an application that queried eUtils to retrieve nephrology-specific biomedical literature citations for a user-definable set of genes. We later augmented the application code to create a general-purpose library that accesses eUtils capability as individual functions that could be combined into user-defined pipelines.ResultsThe toolset’s use is illustrated with an application that serves as a front-end to the library and can be used by non-programmers to construct user-defined pipelines. The operation of the library is illustrated for the literature-surveillance application, which serves as a case-study. An overview of the library is also provided.ConclusionsThe library simplifies use of the eUtils service by operating at a higher level, and also transparently addresses robustness issues that would need to be individually implemented otherwise, such as error recovery and prevention of overloading of the eUtils service.