While large language models (LLMs) have achieved state-of-the-art performance on a wide range of medical question answering (QA) tasks, they still face challenges with hallucinations and outdated knowledge. Retrieval-augmented generation (RAG) is a promising solution and has been widely adopted. However, a RAG system can involve multiple flexible components, and there is a lack of best practices regarding the optimal RAG setting for various medical purposes. To systematically evaluate such systems, we propose the Medical Information Retrieval-Augmented Generation Evaluation (MIRAGE), a first-of-its-kind benchmark including 7,663 questions from five medical QA datasets. Using MIRAGE, we conducted large-scale experiments with over 1.8 trillion prompt tokens on 41 combinations of different corpora, retrievers, and backbone LLMs through the MedRAG toolkit introduced in this work. Overall, MedRAG improves the accuracy of six different LLMs by up to 18% over chain-of-thought prompting, elevating the performance of GPT-3.5 and Mixtral to GPT-4-level. Our results show that the combination of various medical corpora and retrievers achieves the best performance. In addition, we discovered a log-linear scaling property and the “lost-in-the-middle” effects in medical RAG. We believe our comprehensive evaluations can serve as practical guidelines for implementing RAG systems for medicine.

With the growing availability of full-text articles, integrating abstracts and full texts of documents into a unified representation is essential for comprehensive search of scientific literature. However, previous studies have shown that naïvely merging abstracts with full texts of articles does not consistently yield better performance. Balancing the contribution of query terms appearing in the abstract and in sections of different importance in full text articles remains a challenge both with traditional bag-of-words IR approaches and for neural retrieval methods. In this work we establish the connection between the BM25 score of a query term appearing in a section of a full text document and the probability of that document being clicked or identified as relevant. Probability is computed using Pool Adjacent Violators (PAV), an isotonic regression algorithm, providing a maximum likelihood estimate based on the observed data. Using this probabilistic transformation of BM25 scores we show an improved performance on the PubMed Click dataset developed and presented in this study, as well as the 2007 TREC Genomics collection.

Lymph node status plays a pivotal role in the treatment of cancer. The extraction of lymph nodes from radiology text reports enables large-scale training of lymph node detection on MRI. In this work, we first propose an ontology of 41 types of abdominal lymph nodes with a hierarchical relationship. We then introduce an end-to-end approach based on the combination of rules and transformer-based methods to detect these abdominal lymph node mentions and classify their types from the MRI radiology reports. We demonstrate the superior performance of a model fine-tuned on MRI reports using BlueBERT, called MriBERT. We find that MriBERT outperforms the rule-based labeler (0.957 vs 0.644 in micro weighted F1-score) as well as other BERT-based variations (0.913 - 0.928). We make the code and MriBERT publicly available at https://github.com/ncbi-nlp/bluebert, with the hope that this method can facilitate the development of medical report annotators to produce labels from scratch at scale.

The number of unique terms in the scientific literature used to refer to either SARS-CoV-2 or COVID-19 is remarkably large and has continued to increase rapidly despite well-established standardized terms. This high degree of term variation makes high recall identification of these important entities difficult. In this manuscript we present an extensive dictionary of terms used in the literature to refer to SARS-CoV-2 and COVID-19. We use a rule-based approach to iteratively generate new term variants, then locate these variants in a large text corpus. We compare our dictionary to an extensive collection of terminological resources, demonstrating that our resource provides a substantial number of additional terms. We use our dictionary to analyze the usage of SARS-CoV-2 and COVID-19 terms over time and show that the number of unique terms continues to grow rapidly. Our dictionary is freely available at https://github.com/ncbi-nlp/CovidTermVar.

Multi-task learning (MTL) has achieved remarkable success in natural language processing applications. In this work, we study a multi-task learning model with multiple decoders on varieties of biomedical and clinical natural language processing tasks such as text similarity, relation extraction, named entity recognition, and text inference. Our empirical results demonstrate that the MTL fine-tuned models outperform state-of-the-art transformer models (e.g., BERT and its variants) by 2.0% and 1.3% in biomedical and clinical domain adaptation, respectively. Pairwise MTL further demonstrates more details about which tasks can improve or decrease others. This is particularly helpful in the context that researchers are in the hassle of choosing a suitable model for new problems. The code and models are publicly available at https://github.com/ncbi-nlp/bluebert.

Inspired by the success of the General Language Understanding Evaluation benchmark, we introduce the Biomedical Language Understanding Evaluation (BLUE) benchmark to facilitate research in the development of pre-training language representations in the biomedicine domain. The benchmark consists of five tasks with ten datasets that cover both biomedical and clinical texts with different dataset sizes and difficulties. We also evaluate several baselines based on BERT and ELMo and find that the BERT model pre-trained on PubMed abstracts and MIMIC-III clinical notes achieves the best results. We make the datasets, pre-trained models, and codes publicly available at https://github.com/ncbi-nlp/BLUE_Benchmark.

Query auto completion (QAC) systems are a standard part of search engines in industry, helping users formulate their query. Such systems update their suggestions after the user types each character, predicting the user’s intent using various signals – one of the most common being popularity. Recently, deep learning approaches have been proposed for the QAC task, to specifically address the main limitation of previous popularity-based methods: the inability to predict unseen queries. In this work we improve previous methods based on neural language modeling, with the goal of building an end-to-end system. We particularly focus on using real-world data by integrating user information for personalized suggestions when possible. We also make use of time information and study how to increase diversity in the suggestions while studying the impact on scalability. Our empirical results demonstrate a marked improvement on two separate datasets over previous best methods in both accuracy and scalability, making a step towards neural query auto-completion in production search engines.

A search that is targeted at finding a specific document in databases is called a Single Citation search. Single citation searches are particularly important for scholarly databases, such as PubMed, because users are frequently searching for a specific publication. In this work we describe SingleCite, a single citation matching system designed to facilitate user’s search for a specific document. We report on the progress that has been achieved towards building that functionality.

Creating simulated search environments has been of a significant interest in infor-mation retrieval, in both general and bio-medical search domains. Existing collec-tions include modest number of queries and are constructed by manually evaluat-ing retrieval results. In this work we pro-pose leveraging MeSH term assignments for creating synthetic test beds. We select a suitable subset of MeSH terms as queries, and utilize MeSH term assignments as pseudo-relevance rankings for retrieval evaluation. Using well studied retrieval functions, we show that their performance on the proposed data is consistent with similar findings in previous work. We further use the proposed retrieval evaluation framework to better understand how to combine heterogeneous sources of textual information.

State-of-the-art methods for protein-protein interaction (PPI) extraction are primarily feature-based or kernel-based by leveraging lexical and syntactic information. But how to incorporate such knowledge in the recent deep learning methods remains an open question. In this paper, we propose a multichannel dependency-based convolutional neural network model (McDepCNN). It applies one channel to the embedding vector of each word in the sentence, and another channel to the embedding vector of the head of the corresponding word. Therefore, the model can use richer information obtained from different channels. Experiments on two public benchmarking datasets, AIMed and BioInfer, demonstrate that McDepCNN provides up to 6% F1-score improvement over rich feature-based methods and single-kernel methods. In addition, McDepCNN achieves 24.4% relative improvement in F1-score over the state-of-the-art methods on cross-corpus evaluation and 12% improvement in F1-score over kernel-based methods on “difficult” instances. These results suggest that McDepCNN generalizes more easily over different corpora, and is capable of capturing long distance features in the sentences.

The Precision Medicine Track in BioCre-ative VI aims to bring together the Bi-oNLP community for a novel challenge focused on mining the biomedical litera-ture in search of mutations and protein-protein interactions (PPI). In order to support this track with an effective train-ing dataset with limited curator time, the track organizers carefully reviewed Pub-Med articles from two different sources: curated public PPI databases, and the re-sults of state-of-the-art public text mining tools. We detail here the data collection, manual review and annotation process and describe this training corpus charac-teristics. We also describe a corpus per-formance baseline. This analysis will provide useful information to developers and researchers for comparing and devel-oping innovative text mining approaches for the BioCreative VI challenge and other Precision Medicine related applica-tions.

We describe a Deep Learning approach to modeling the relevance of a document’s text to a query, applied to biomedical literature. Instead of mapping each document and query to a common semantic space, we compute a variable-length difference vector between the query and document which is then passed through a deep convolution stage followed by a deep regression network to produce the estimated probability of the document’s relevance to the query. Despite the small amount of training data, this approach produces a more robust predictor than computing similarities between semantic vector representations of the query and document, and also results in significant improvements over traditional IR text factors. In the future, we plan to explore its application in improving PubMed search.