Exploiting and assessing multi-source data for supervised biomedical named entity recognition

Extracting knowledge networks from plant scientific literature: potato tuber flesh color as an exemplary trait

BACKGROUND

Scientific literature carries a wealth of information crucial for research, but only a fraction of it is present as structured information in databases and therefore can be analyzed using traditional data analysis tools. Natural language processing (NLP) is often and successfully employed to support humans by distilling relevant information from large corpora of free text and structuring it in a way that lends itself to further computational analyses. For this pilot, we developed a pipeline that uses NLP on biological literature to produce knowledge networks. We focused on the flesh color of potato, a well-studied trait with known associations, and we investigated whether these knowledge networks can assist us in formulating new hypotheses on the underlying biological processes.

RESULTS

We trained an NLP model based on a manually annotated corpus of 34 full-text potato articles, to recognize relevant biological entities and relationships between them in text (genes, proteins, metabolites and traits). This model detected the number of biological entities with a precision of 97.65% and a recall of 88.91% on the training set. We conducted a time series analysis on 4023 PubMed abstract of plant genetics-based articles which focus on 4 major Solanaceous crops (tomato, potato, eggplant and capsicum), to determine that the networks contained both previously known and contemporaneously unknown leads to subsequently discovered biological phenomena relating to flesh color. A novel time-based analysis of these networks indicates a connection between our trait and a candidate gene (zeaxanthin epoxidase) already two years prior to explicit statements of that connection in the literature.

CONCLUSIONS

Our time-based analysis indicates that network-assisted hypothesis generation shows promise for knowledge discovery, data integration and hypothesis generation in scientific research.

Named entity recognition on bio-medical literature documents using hybrid based approach

There have been many changes in the medical field due to technological advances. The progression in technologies provides lot of opportunities to extract valuable insights from huge amount of unstructured data. The literature documents published by the researchers in medical domain consists enormous amount of knowledge. Many organizations are involving in retrieving the hidden information from the literature documents. Extracting the drug names, diseases, symptoms, route of administration, species and dosage forms from the textual document is an easy task due to the innovation of technologies in the Natural Language Processing. In this article, a new hybrid based approach is proposed to identify named entity from the medical literature documents. New dictionary has been built for route of administration, dosage forms and symptoms to annotate the entities in the medical documents. The annotated entities are trained by the blank Spacy machine learning model. The trained model provide a decent accuracy when compared with the existing model. The hybrid model is validated with the dictionary and human (optional)to calculate the confusion matrix. It is able to identify more entities than the prevailing model. The average F1 score for five entities of the proposed hybrid based approach 73.79%.

Towards reliable named entity recognition in the biomedical domain

MOTIVATION

Automatic biomedical named entity recognition (BioNER) is a key task in biomedical information extraction. For some time, state-of-the-art BioNER has been dominated by machine learning methods, particularly conditional random fields (CRFs), with a recent focus on deep learning. However, recent work has suggested that the high performance of CRFs for BioNER may not generalize to corpora other than the one it was trained on. In our analysis, we find that a popular deep learning-based approach to BioNER, known as bidirectional long short-term memory network-conditional random field (BiLSTM-CRF), is correspondingly poor at generalizing. To address this, we evaluate three modifications of BiLSTM-CRF for BioNER to improve generalization: improved regularization via variational dropout, transfer learning and multi-task learning.

RESULTS

We measure the effect that each strategy has when training/testing on the same corpus ('in-corpus' performance) and when training on one corpus and evaluating on another ('out-of-corpus' performance), our measure of the model's ability to generalize. We found that variational dropout improves out-of-corpus performance by an average of 4.62%, transfer learning by 6.48% and multi-task learning by 8.42%. The maximal increase we identified combines multi-task learning and variational dropout, which boosts out-of-corpus performance by 10.75%. Furthermore, we make available a new open-source tool, called Saber that implements our best BioNER models.

AVAILABILITY AND IMPLEMENTATION

Source code for our biomedical IE tool is available at https://github.com/BaderLab/saber. Corpora and other resources used in this study are available at https://github.com/BaderLab/Towards-reliable-BioNER.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Flagging incorrect nucleotide sequence reagents in biomedical papers: To what extent does the leading publication format impede automatic error detection?

In an idealised vision of science the scientific literature is error-free. Errors reported during peer review are supposed to be corrected prior to publication, as further research establishes new knowledge based on the body of literature. It happens, however, that errors pass through peer review, and a minority of cases errata and retractions follow. Automated screening software can be applied to detect errors in manuscripts and publications. The contribution of this paper is twofold. First, we designed the erroneous reagent checking () benchmark to assess the accuracy of fact-checkers screening biomedical publications for dubious mentions of nucleotide sequence reagents. It comes with a test collection comprised of 1679 nucleotide sequence reagents that were curated by biomedical experts. Second, we benchmarked our own screening software called Seek&Blastn with three input formats to assess the extent of performance loss when operating on various publication formats. Our findings stress the superiority of markup formats (a 79% detection rate on XML and HTML) over the prominent PDF format (a 69% detection rate at most) regarding an error flagging task. This is the first published baseline on error detection involving reagents reported in biomedical scientific publications. The benchmark is designed to facilitate the development and validation of software bricks to enhance the reliability of the peer review process.

Biomedical named entity recognition using deep neural networks with contextual information

BACKGROUND

In biomedical text mining, named entity recognition (NER) is an important task used to extract information from biomedical articles. Previously proposed methods for NER are dictionary- or rule-based methods and machine learning approaches. However, these traditional approaches are heavily reliant on large-scale dictionaries, target-specific rules, or well-constructed corpora. These methods to NER have been superseded by the deep learning-based approach that is independent of hand-crafted features. However, although such methods of NER employ additional conditional random fields (CRF) to capture important correlations between neighboring labels, they often do not incorporate all the contextual information from text into the deep learning layers.

RESULTS

We propose herein an NER system for biomedical entities by incorporating n-grams with bi-directional long short-term memory (BiLSTM) and CRF; this system is referred to as a contextual long short-term memory networks with CRF (CLSTM). We assess the CLSTM model on three corpora: the disease corpus of the National Center for Biotechnology Information (NCBI), the BioCreative II Gene Mention corpus (GM), and the BioCreative V Chemical Disease Relation corpus (CDR). Our framework was compared with several deep learning approaches, such as BiLSTM, BiLSTM with CRF, GRAM-CNN, and BERT. On the NCBI corpus, our model recorded an F-score of 85.68% for the NER of diseases, showing an improvement of 1.50% over previous methods. Moreover, although BERT used transfer learning by incorporating more than 2.5 billion words, our system showed similar performance with BERT with an F-scores of 81.44% for gene NER on the GM corpus and a outperformed F-score of 86.44% for the NER of chemicals and diseases on the CDR corpus. We conclude that our method significantly improves performance on biomedical NER tasks.

CONCLUSION

The proposed approach is robust in recognizing biological entities in text.

HyperFoods: Machine intelligent mapping of cancer-beating molecules in foods

Recent data indicate that up-to 30–40% of cancers can be prevented by dietary and lifestyle measures alone. Herein, we introduce a unique network-based machine learning platform to identify putative food-based cancer-beating molecules. These have been identified through their molecular biological network commonality with clinically approved anti-cancer therapies. A machine-learning algorithm of random walks on graphs (operating within the supercomputing DreamLab platform) was used to simulate drug actions on human interactome networks to obtain genome-wide activity profiles of 1962 approved drugs (199 of which were classified as “anti-cancer” with their primary indications). A supervised approach was employed to predict cancer-beating molecules using these ‘learned’ interactome activity profiles. The validated model performance predicted anti-cancer therapeutics with classification accuracy of 84–90%. A comprehensive database of 7962 bioactive molecules within foods was fed into the model, which predicted 110 cancer-beating molecules (defined by anti-cancer drug likeness threshold of >70%) with expected capacity comparable to clinically approved anti-cancer drugs from a variety of chemical classes including flavonoids, terpenoids, and polyphenols. This in turn was used to construct a ‘food map’ with anti-cancer potential of each ingredient defined by the number of cancer-beating molecules found therein. Our analysis underpins the design of next-generation cancer preventative and therapeutic nutrition strategies.

Precision Medicine in Pancreatic Disease-Knowledge Gaps and Research Opportunities: Summary of a National Institute of Diabetes and Digestive and Kidney Diseases Workshop

A workshop on research gaps and opportunities for Precision Medicine in Pancreatic Disease was sponsored by the National Institute of Diabetes and Digestive Kidney Diseases on July 24, 2019, in Pittsburgh. The workshop included an overview lecture on precision medicine in cancer and 4 sessions: (1) general considerations for the application of bioinformatics and artificial intelligence; (2) omics, the combination of risk factors and biomarkers; (3) precision imaging; and (4) gaps, barriers, and needs to move from precision to personalized medicine for pancreatic disease. Current precision medicine approaches and tools were reviewed, and participants identified knowledge gaps and research needs that hinder bringing precision medicine to pancreatic diseases. Most critical were (a) multicenter efforts to collect large-scale patient data sets from multiple data streams in the context of environmental and social factors; (b) new information systems that can collect, annotate, and quantify data to inform disease mechanisms; (c) novel prospective clinical trial designs to test and improve therapies; and (d) a framework for measuring and assessing the value of proposed approaches to the health care system. With these advances, precision medicine can identify patients early in the course of their pancreatic disease and prevent progression to chronic or fatal illness.