Background: The growing recognition of the microbiome’s impact on human health and well-being has prompted extensive research into discovering the links between microbiome dysbiosis and disease (healthy) states. However, this valuable information is scattered in unstructured form within biomedical literature. The structured extraction and qualification of microbe-disease interactions are important. In parallel, recent advancements in deep-learning-based natural language processing algorithms have revolutionized language-related tasks such as ours. This study aims to leverage state-of-the-art deep-learning language models to extract microbe-disease relationships from biomedical literature. Results: In this study, we first evaluate multiple pre-trained large language models within a zero-shot or few-shot learning context. In this setting, the models performed poorly out of the box, emphasizing the need for domain-specific fine-tuning of these language models. Subsequently, we fine-tune multiple language models (specifically, GPT-3, BioGPT, BioMedLM, BERT, BioMegatron, PubMedBERT, BioClinicalBERT, and BioLinkBERT) using labeled training data and evaluate their performance. Our experimental results demonstrate the state-of-the-art performance of these fine-tuned models ( specifically GPT-3, BioMedLM, and BioLinkBERT), achieving an average F1 score, precision, and recall of over $$>0.8$$ compared to the previous best of 0.74. Conclusion: Overall, this study establishes that pre-trained language models excel as transfer learners when fine-tuned with domain and problem-specific data, enabling them to achieve state-of-the-art results even with limited training data for extracting microbiome-disease interactions from scientific publications.
18 August 2020 in IEEE Transactions on Knowledge and Data Engineering
In this paper, we study an automatic hypothesis generation (HG) problem, which refers to the discovery of meaningful implicit connections between scientific terms, including but not limited to diseases, chemicals, drugs, and genes extracted from databases of biomedical publications. Most prior studies of this problem focused on the use of static information of terms and largely ignored the temporal dynamics of scientific term relations. Even when the dynamics were considered in a few recent studies, they learned the representations for the scientific terms, rather than focusing on the term-pair relations. Since the HG problem is to predict term-pair connections, it is not enough to know with whom the terms are connected; it is more important to know how the connections have been formed (in a dynamic process). We formulate this HG problem as a future connectivity prediction in a dynamic attributed graph. The key is to capture the temporal evolution of node-pair (term-pair) relations. We propose an inductive edge (node-pair) embedding method named T-PAIR, utilizing both the graphical structure and node attribute to encode the temporal node-pair relationship. We demonstrate the efficiency of the proposed model on three real-world datasets constructed from biomedical publications in the transductive and inductive settings.
Background: Maintaining good communication and engagement between people with dementia and their caregivers is a major challenge in dementia care. Cognitive stimulation is a psychosocial intervention that supports communication and engagement, and several digital applications for cognitive stimulation have been developed. Personalization is an important factor for obtaining sustainable benefits, but the time and effort required to personalize and optimize applications often makes them difficult for routine use by nonspecialist caregivers and families. Although artificial intelligence (AI) has great potential to support automation of the personalization process, its use is largely unexplored because of the lack of suitable data from which to develop and train machine learning models. Objective: This pilot study aims to evaluate a digital application called Aikomi in Japanese care homes for its potential to (1) create and deliver personalized cognitive stimulation programs to promote communication and engagement between people with dementia and usual care staff and (2) capture meaningful personalized data suitable for the development of AI systems. Methods: A modular technology platform was developed and used to create personalized programs for 15 people with dementia living in 4 residential care facilities in Japan with the cooperation of a family member or care staff. A single intervention with the program was conducted with the person with dementia together with a care staff member, and for some participants, smell stimulation was provided using selected smell sticks in conjunction with the digital program. All sessions were recorded using a video camera, and the combined personalized data obtained by the platform were analyzed. Results: Most people with dementia (10/15, 67%) showed high levels of engagement (>40 on Engagement of a Person with Dementia Scale), and there were no incidences of negative reactions toward the programs. Care staff reported that some participants showed extended concentration and spontaneous communication while using Aikomi, which was not their usual behavior. Smell stimulation promoted engagement for some participants even when they were unable to identify the smell. No changes in well-being were observed following the intervention according to the Mental Function Impairment Scale. The level of response to each type of content in the stimulation program varied greatly according to the person with dementia, and personalized data captured by the Aikomi platform enabled understanding of correlations between stimulation content and responses for each participant. Conclusions: This study suggests that the Aikomi digital application is acceptable for use by persons with dementia and care staff and may have the potential to promote communication and engagement. The platform captures personalized data, which can provide suitable input for machine learning. Further investigation of Aikomi will be conducted to develop AI systems and create personalized digital cognitive stimulation applications that can be easily used by nonspecialist caregivers.
Protocol standardization and sharing are crucial for reproducibility in life sciences. In spite of numerous efforts for standardized protocol description, adherence to these standards in literature remains largely inconsistent. Curation of protocols are especially challenging due to the labor intensive process, requiring expert domain knowledge of each experimental procedure. Recent advancements in Large Language Models (LLMs) offer a promising solution to interpret and curate knowledge from complex scientific literature. In this work, we develop ProtoCode, a tool leveraging fine-tune LLMs to curate protocols into intermediate representation formats which can be interpretable by both human and machine interfaces. Our proof-of-concept, focused on polymerase chain reaction (PCR) protocols, retrieves information from PCR protocols at an accuracy ranging 69-100 % depending on the information content. In all tested protocols, we demonstrate that ProtoCode successfully converts literature-based protocols into correct operational files for multiple thermal cycler systems. In conclusion, ProtoCode can alleviate labor intensive curation and standardization of life science protocols to enhance research reproducibility by providing a reliable, automated means to process and standardize protocols. ProtoCode is freely available as a web server at https://curation.taxila.io/ProtoCode/.
Text mining has been shown to be an auxiliary but key driver for modeling, data harmonization, and interpretation in bio-medicine. Scientific literature holds a wealth of information and embodies cumulative knowledge and remains the core basis on which mechanistic pathways, molecular databases, and models are built and refined. Text mining provides the necessary tools to automatically harness the potential of text. In this study, we show the potential of large-scale text mining for deriving novel insights, with a focus on the growing field of microbiome. We first collected the complete set of abstracts relevant to the microbiome from PubMed and used our text mining and intelligence platform Taxila for analysis. We drive the usefulness of text mining using two case studies. First, we analyze the geographical distribution of research and study locations for the field of microbiome by extracting geo mentions from text. Using this analysis, we were able to draw useful insights on the state of research in microbiome w. r.t geographical distributions and economic drivers. Next, to understand the relationships between diseases, microbiome, and food which are central to the field, we construct semantic relationship networks between these different concepts central to the field of microbiome. We show how such networks can be useful to derive useful insight with no prior knowledge encoded.
Over the last few years Literature-based Discovery (LBD) has regained popularity as a means to enhance the scientific research process. The resurgent interest has spurred the development of supervised and semi-supervised machine learning models aimed at making previously implicit connections between scientific concepts/entities within often extensive bodies of literature explicit—i.e., suggesting novel scientific hypotheses. In doing so, understanding the temporally evolving interactions between these entities can provide valuable information for predicting the future development of entity relationships. However, existing methods often underutilize the latent information embedded in the temporal aspects of the interaction data. Motivated by applications in the food domain—where we aim to connect nutritional information with health-related benefits—we address the hypothesis-generation problem using a temporal graph-based approach. Given that hypothesis generation involves predicting future (i.e., still to be discovered) entity connections, in our view the ability to capture the dynamic evolution of connections over time is pivotal for a robust model. To address this, we introduce THiGER, a novel batch contrastive temporal node-pair embedding method. THiGER excels in providing a more expressive node-pair encoding by effectively harnessing node-pair relationships. Furthermore, we present THiGER-A, an incremental training approach that incorporates an active curriculum learning strategy to mitigate label bias arising from unobserved connections. By progressively training on increasingly challenging and high-utility samples, our approach significantly enhances the performance of the embedding model. Empirical validation of our proposed method demonstrates its effectiveness on established temporal-graph benchmark datasets, as well as on real-world datasets within the food domain.