Association measures for estimating semantic similarity and relatedness between biomedical concepts

Improved biomedical word embeddings in the transformer era

BACKGROUND

Recent natural language processing (NLP) research is dominated by neural network methods that employ word embeddings as basic building blocks. Pre-training with neural methods that capture local and global distributional properties (e.g., skip-gram, GLoVE) using free text corpora is often used to embed both words and concepts. Pre-trained embeddings are typically leveraged in downstream tasks using various neural architectures that are designed to optimize task-specific objectives that might further tune such embeddings.

OBJECTIVE

Despite advances in contextualized language model based embeddings, static word embeddings still form an essential starting point in BioNLP research and applications. They are useful in low resource settings and in lexical semantics studies. Our main goal is to build improved biomedical word embeddings and make them publicly available for downstream applications.

METHODS

We jointly learn word and concept embeddings by first using the skip-gram method and further fine-tuning them with correlational information manifesting in co-occurring Medical Subject Heading (MeSH) concepts in biomedical citations. This fine-tuning is accomplished with the transformer-based BERT architecture in the two-sentence input mode with a classification objective that captures MeSH pair co-occurrence. We conduct evaluations of these tuned static embeddings using multiple datasets for word relatedness developed by previous efforts.

RESULTS

Both in qualitative and quantitative evaluations we demonstrate that our methods produce improved biomedical embeddings in comparison with other static embedding efforts. Without selectively culling concepts and terms (as was pursued by previous efforts), we believe we offer the most exhaustive evaluation of biomedical embeddings to date with clear performance improvements across the board.

CONCLUSION

We repurposed a transformer architecture (typically used to generate dynamic embeddings) to improve static biomedical word embeddings using concept correlations. We provide our code and embeddings for public use for downstream applications and research endeavors: https://github.com/bionlproc/BERT-CRel-Embeddings.

Using Literature Based Discovery to Gain Insights Into the Metabolomic Processes of Cardiac Arrest

In this paper, we describe how we applied LBD techniques to discover lecithin cholesterol acyltransferase (LCAT) as a druggable target for cardiac arrest. We fully describe our process which includes the use of high-throughput metabolomic analysis to identify metabolites significantly related to cardiac arrest, and how we used LBD to gain insights into how these metabolites relate to cardiac arrest. These insights lead to our proposal (for the first time) of LCAT as a druggable target; the effects of which are supported by in vivo studies which were brought forth by this work. Metabolites are the end product of many biochemical pathways within the human body. Observed changes in metabolite levels are indicative of changes in these pathways, and provide valuable insights toward the cause, progression, and treatment of diseases. Following cardiac arrest, we observed changes in metabolite levels pre- and post-resuscitation. We used LBD to help discover diseases implicitly linked via these metabolites of interest. Results of LBD indicated a strong link between Fish Eye disease and cardiac arrest. Since fish eye disease is characterized by an LCAT deficiency, it began an investigation into the effects of LCAT and cardiac arrest survival. In the investigation, we found that decreased LCAT activity may increase cardiac arrest survival rates by increasing ω-3 polyunsaturated fatty acid availability in circulation. We verified the effects of ω-3 polyunsaturated fatty acids on increasing survival rate following cardiac arrest via in vivo with rat models.

A Neuro-ontology for the neurological examination

BACKGROUND

The use of clinical data in electronic health records for machine-learning or data analytics depends on the conversion of free text into machine-readable codes. We have examined the feasibility of capturing the neurological examination as machine-readable codes based on UMLS Metathesaurus concepts.

METHODS

We created a target ontology for capturing the neurological examination using 1100 concepts from the UMLS Metathesaurus. We created a dataset of 2386 test-phrases based on 419 published neurological cases. We then mapped the test-phrases to the target ontology.

RESULTS

We were able to map all of the 2386 test-phrases to 601 unique UMLS concepts. A neurological examination ontology with 1100 concepts has sufficient breadth and depth of coverage to encode all of the neurologic concepts derived from the 419 test cases. Using only pre-coordinated concepts, component ontologies of the UMLS, such as HPO, SNOMED CT, and OMIM, do not have adequate depth and breadth of coverage to encode the complexity of the neurological examination.

CONCLUSION

An ontology based on a subset of UMLS has sufficient breadth and depth of coverage to convert deficits from the neurological examination into machine-readable codes using pre-coordinated concepts. The use of a small subset of UMLS concepts for a neurological examination ontology offers the advantage of improved manageability as well as the opportunity to curate the hierarchy and subsumption relationships.

Measuring similarity in co-occurrence data using ego-networks

The co-occurrence association is widely observed in many empirical data. Mining the information in co-occurrence data is essential for advancing our understanding of systems such as social networks, ecosystems, and brain networks. Measuring similarity of entities is one of the important tasks, which can usually be achieved using a network-based approach. Here, we show that traditional methods based on the aggregated network can bring unwanted indirect relationships. To cope with this issue, we propose a similarity measure based on the ego network of each entity, which effectively considers the change of an entity's centrality from one ego network to another. The index proposed is easy to calculate and has a clear physical meaning. Using two different data sets, we compare the new index with other existing ones. We find that the new index outperforms the traditional network-based similarity measures, and it can sometimes surpass the embedding method. In the meanwhile, the measure by the new index is weakly correlated with those by other methods, hence providing a different dimension to quantify similarities in co-occurrence data. Altogether, our work makes an extension in the network-based similarity measure and can be potentially applied in several related tasks.

Indirect association and ranking hypotheses for literature based discovery

BACKGROUND

Literature Based Discovery (LBD) produces more potential hypotheses than can be manually reviewed, making automatically ranking these hypotheses critical. In this paper, we introduce the indirect association measures of Linking Term Association (LTA), Minimum Weight Association (MWA), and Shared B to C Set Association (SBC), and compare them to Linking Set Association (LSA), concept embeddings vector cosine, Linking Term Count (LTC), and direct co-occurrence vector cosine. Our proposed indirect association measures extend traditional association measures to quantify indirect rather than direct associations while preserving valuable statistical properties.

RESULTS

We perform a comparison between several different hypothesis ranking methods for LBD, and compare them against our proposed indirect association measures. We intrinsically evaluate each method's performance using its ability to estimate semantic relatedness on standard evaluation datasets. We extrinsically evaluate each method's ability to rank hypotheses in LBD using a time-slicing dataset based on co-occurrence information, and another time-slicing dataset based on SemRep extracted-relationships. Precision and recall curves are generated by ranking term pairs and applying a threshold at each rank.

CONCLUSIONS

Results differ depending on the evaluation methods and datasets, but it is unclear if this is a result of biases in the evaluation datasets or if one method is truly better than another. We conclude that LTC and SBC are the best suited methods for hypothesis ranking in LBD, but there is value in having a variety of methods to choose from.

Measuring semantic similarity of clinical trial outcomes using deep pre-trained language representations

BACKGROUND

Outcomes are variables monitored during a clinical trial to assess the impact of an intervention on humans' health.Automatic assessment of semantic similarity of trial outcomes is required for a number of tasks, such as detection of outcome switching (unjustified changes of pre-defined outcomes of a trial) and implementation of Core Outcome Sets (minimal sets of outcomes that should be reported in a particular medical domain).

OBJECTIVE

We aimed at building an algorithm for assessing semantic similarity of pairs of primary and reported outcomes.We focused on approaches that do not require manually curated domain-specific resources such as ontologies and thesauri.

METHODS

We tested several approaches, including single measures of similarity (based on strings, stems and lemmas, paths and distances in an ontology, and vector representations of phrases), classifiers using a combination of single measures as features, and a deep learning approach that consists in fine-tuning pre-trained deep language representations.We tested language models provided by BERT (trained on general-domain texts), BioBERT and SciBERT (trained on biomedical and scientific texts, respectively).We explored the possibility of improving the results by taking into account the variants for referring to an outcome (e.g.the use of a measurement tool name instead on the outcome name; the use of abbreviations).We release an open corpus with annotation for similarity of pairs of outcomes.

RESULTS

Classifiers using a combination of single measures as features outperformed the single measures, while deep learning algorithms using BioBERT and SciBERT models outperformed the classifiers.BioBERT reached the best F-measure of 89.75%.The addition of variants of outcomes did not improve the results for the best-performing single measures nor for the classifiers, but it improved the performance of deep learning algorithms: BioBERT achieved an F-measure of93.38%.

CONCLUSIONS

Deep learning approaches using pre-trained language representations outperformed other approaches for similarity assessment of trial outcomes, without relying on any manually curated domain-specific resources (ontologies and other lexical resources). Addition of variants of outcomes further improved the performance of deep learning algorithms.