A curated transcriptome dataset collection to investigate the functional programming of human hematopoietic cells in early life

Using semantic search to find publicly available gene-expression datasets

Millions of high-throughput, molecular datasets have been shared in public repositories. have been shared in public repositories. Researchers can reuse such data to validate their own findings and explore novel questions. A frequent goal is to find multiple datasets that address similar research topics and to either combine them directly or integrate inferences from them. However, a major challenge is finding relevant datasets due to the vast number of candidates, inconsistencies in their descriptions, and a lack of semantic annotations. This challenge is first among the FAIR principles for scientific data. Here we focus on dataset discovery within Gene Expression Omnibus (GEO), a repository containing 100,000s of data series. GEO supports queries based on keywords, ontology terms, and other annotations. However, reviewing these results is time-consuming and tedious, and it often misses relevant datasets. We hypothesized that language models could address this problem by summarizing dataset descriptions as numeric representations (embeddings). Assuming a researcher has previously found some relevant datasets, we evaluated the potential to find additional relevant datasets. For six human medical conditions, we used 30 models to generate embeddings for datasets that human curators had previously associated with the conditions and identified other datasets with the most similar descriptions. This approach was often, but not always, more effective than GEO's search engine. Our top-performing models were trained on general corpora, used contrastive-learning strategies, and used relatively large embeddings. Our findings suggest that language models have the potential to improve dataset discovery, perhaps in combination with existing search tools.

Harnessing large language models (LLMs) for candidate gene prioritization and selection

BACKGROUND

Feature selection is a critical step for translating advances afforded by systems-scale molecular profiling into actionable clinical insights. While data-driven methods are commonly utilized for selecting candidate genes, knowledge-driven methods must contend with the challenge of efficiently sifting through extensive volumes of biomedical information. This work aimed to assess the utility of large language models (LLMs) for knowledge-driven gene prioritization and selection.

METHODS

In this proof of concept, we focused on 11 blood transcriptional modules associated with an Erythroid cells signature. We evaluated four leading LLMs across multiple tasks. Next, we established a workflow leveraging LLMs. The steps consisted of: (1) Selecting one of the 11 modules; (2) Identifying functional convergences among constituent genes using the LLMs; (3) Scoring candidate genes across six criteria capturing the gene's biological and clinical relevance; (4) Prioritizing candidate genes and summarizing justifications; (5) Fact-checking justifications and identifying supporting references; (6) Selecting a top candidate gene based on validated scoring justifications; and (7) Factoring in transcriptome profiling data to finalize the selection of the top candidate gene.

RESULTS

Of the four LLMs evaluated, OpenAI's GPT-4 and Anthropic's Claude demonstrated the best performance and were chosen for the implementation of the candidate gene prioritization and selection workflow. This workflow was run in parallel for each of the 11 erythroid cell modules by participants in a data mining workshop. Module M9.2 served as an illustrative use case. The 30 candidate genes forming this module were assessed, and the top five scoring genes were identified as BCL2L1, ALAS2, SLC4A1, CA1, and FECH. Researchers carefully fact-checked the summarized scoring justifications, after which the LLMs were prompted to select a top candidate based on this information. GPT-4 initially chose BCL2L1, while Claude selected ALAS2. When transcriptional profiling data from three reference datasets were provided for additional context, GPT-4 revised its initial choice to ALAS2, whereas Claude reaffirmed its original selection for this module.

CONCLUSIONS

Taken together, our findings highlight the ability of LLMs to prioritize candidate genes with minimal human intervention. This suggests the potential of this technology to boost productivity, especially for tasks that require leveraging extensive biomedical knowledge.

Organizing training workshops on gene literature retrieval, profiling, and visualization for early career researchers

Early-career researchers must acquire the skills necessary to effectively search and extract information from biomedical literature. This ability is for instance crucial for evaluating the novelty of experimental results, and assessing potential publishing opportunities. Given the rapidly growing volume of publications in the field of biomedical research, new systematic approaches need to be devised and adopted for the retrieval and curation of literature relevant to a specific theme. In this context, we present a hands-on training curriculum aimed at retrieval, profiling, and visualization of literature associated with a given topic. The curriculum was implemented in a workshop in January 2021. Here we provide supporting material and step-by-step implementation guidelines with the ISG15 gene literature serving as an illustrative use case. Workshop participants can learn several skills, including: 1) building and troubleshoot PubMed queries in order to retrieve the literature associated with a gene of interest; 2) identifying key concepts relevant to given themes (such as cell types, diseases, and biological processes); 3) measuring the prevalence of these concepts in the gene literature; 4) extracting key information from relevant articles, and 5) developing a background section or summary on the basis of this information. Finally, trainees can learn to consolidate the structured information captured through this process for presentation via an interactive web application.

Organizing gene literature retrieval, profiling, and visualization training workshops for early career researchers

Developing the skills needed to effectively search and extract information from biomedical literature is essential for early-career researchers. It is, for instance, on this basis that the novelty of experimental results, and therefore publishing opportunities, can be evaluated. Given the unprecedented volume of publications in the field of biomedical research, new systematic approaches need to be devised and adopted for the retrieval and curation of literature relevant to a specific theme. Here we describe a hands-on training curriculum aimed at retrieval, profiling, and visualization of literature associated with a given topic. This curriculum was implemented in a workshop in January 2021. We provide supporting material and step-by-step implementation guidelines with the ISG15 gene literature serving as an illustrative use case. Through participation in such a workshop, trainees can learn: 1) to build and troubleshoot PubMed queries in order to retrieve the literature associated with a gene of interest; 2) to identify key concepts relevant to given themes (such as cell types, diseases, and biological processes); 3) to measure the prevalence of these concepts in the gene literature; 4) to extract key information from relevant articles, and 5) to develop a background section or summary on the basis of this information. Finally, trainees can learn to consolidate the structured information captured through this process for presentation via an interactive web application.

Long-Chain Acyl-CoA Synthetase 1 Role in Sepsis and Immunity: Perspectives From a Parallel Review of Public Transcriptome Datasets and of the Literature

A potential role for the long-chain acyl-CoA synthetase family member 1 (ACSL1) in the immunobiology of sepsis was explored during a hands-on training workshop. Participants first assessed the robustness of the potential gap in biomedical knowledge identified via an initial screen of public transcriptome data and of the literature associated with ACSL1. Increase in ACSL1 transcript abundance during sepsis was confirmed in several independent datasets. Querying the ACSL1 literature also confirmed the absence of reports associating ACSL1 with sepsis. Inferences drawn from both the literature (via indirect associations) and public transcriptome data (via correlation) point to the likely participation of ACSL1 and ACSL4, another family member, in inflammasome activation in neutrophils during sepsis. Furthermore, available clinical data indicate that levels of ACSL1 and ACSL4 induction was significantly higher in fatal cases of sepsis. This denotes potential translational relevance and is consistent with involvement in pathways driving potentially deleterious systemic inflammation. Finally, while ACSL1 expression was induced in blood in vitro by a wide range of pathogen-derived factors as well as TNF, induction of ACSL4 appeared restricted to flagellated bacteria and pathogen-derived TLR5 agonists and IFNG. Taken together, this joint review of public literature and omics data records points to two members of the acyl-CoA synthetase family potentially playing a role in inflammasome activation in neutrophils. Translational relevance of these observations in the context of sepsis and other inflammatory conditions remain to be investigated.

A curated transcriptome dataset collection to investigate the blood transcriptional response to viral respiratory tract infection and vaccination

The human immune defense mechanisms and factors associated with good versus poor health outcomes following viral respiratory tract infections (VRTI), as well as correlates of protection following vaccination against respiratory viruses, remain incompletely understood. To shed further light into these mechanisms, a number of systems-scale studies have been conducted to measure transcriptional changes in blood leukocytes of either naturally or experimentally infected individuals, or in individual’s post-vaccination. Here we are making available a public repository, for research investigators for interpretation, a collection of transcriptome datasets obtained from human whole blood and peripheral blood mononuclear cells (PBMC) to investigate the transcriptional responses following viral respiratory tract infection or vaccination against respiratory viruses. In total, Thirty one31 datasets, associated to viral respiratory tract infections and their related vaccination studies, were identified and retrieved from the NCBI Gene Expression Omnibus (GEO) and loaded in a custom web application designed for interactive query and visualization of integrated large-scale data. Quality control checks, using relevant biological markers, were performed. Multiple sample groupings and rank lists were created to facilitate dataset query and interpretation. Via this interface, users can generate web links to customized graphical views, which may be subsequently inserted into manuscripts to report novel findings. The GXB tool enables browsing of a single gene across projects, providing new perspectives on the role of a given molecule across biological systems in the diagnostic and prognostic following VRTI but also in identifying new correlates of protection. This dataset collection is available at: http://vri1.gxbsidra.org/dm3/geneBrowser/list.

A curated transcriptome dataset collection to investigate inborn errors of immunity

Primary immunodeficiencies (PIDs) are a heterogeneous group of inherited disorders, frequently caused by loss-of-function and less commonly by gain-of-function mutations, which can result in susceptibility to a broad or a very narrow range of infections but also in inflammatory, allergic or malignant diseases. Owing to the wide range in clinical manifestations and variability in penetrance and expressivity, there is an urgent need to better understand the underlying molecular, cellular and immunological phenotypes in PID patients in order to improve clinical diagnosis and management. Here we have compiled a manually curated collection of public transcriptome datasets mainly obtained from human whole blood, peripheral blood mononuclear cells (PBMCs) or fibroblasts of patients with PIDs and of control subjects for subsequent meta-analysis, query and interpretation. A total of eighteen (18) datasets derived from studies of PID patients were identified and retrieved from the NCBI Gene Expression Omnibus (GEO) database and loaded in GXB, a custom web application designed for interactive query and visualization of integrated large-scale data. The dataset collection includes samples from well characterized PID patients that were stimulated ex vivo under a variety of conditions to assess the molecular consequences of the underlying, naturally occurring gene defects on a genome-wide scale. Multiple sample groupings and rank lists were generated to facilitate comparisons of the transcriptional responses between different PID patients and control subjects. The GXB tool enables browsing of a single transcript across studies, thereby providing new perspectives on the role of a given molecule across biological systems and PID patients. This dataset collection is available at http://pid.gxbsidra.org/dm3/geneBrowser/list.

A curated transcriptome dataset collection to investigate inborn errors of immunity

Primary immunodeficiencies (PIDs) are a heterogeneous group of inherited disorders, frequently caused by loss-of-function and less commonly by gain-of-function mutations, which can result in susceptibility to a broad or a very narrow range of infections but also in inflammatory, allergic or malignant diseases. Owing to the wide range in clinical manifestations and variability in penetrance and expressivity, there is an urgent need to better understand the underlying molecular, cellular and immunological phenotypes in PID patients in order to improve clinical diagnosis and management. Here we have compiled a manually curated collection of public transcriptome datasets mainly obtained from human whole blood, peripheral blood mononuclear cells (PBMCs) or fibroblasts of patients with PIDs and of control subjects for subsequent meta-analysis, query and interpretation. A total of eighteen (18) datasets derived from studies of PID patients were identified and retrieved from the NCBI Gene Expression Omnibus (GEO) database and loaded in GXB, a custom web application designed for interactive query and visualization of integrated large-scale data. The dataset collection includes samples from well characterized PID patients that were stimulated ex vivo under a variety of conditions to assess the molecular consequences of the underlying, naturally occurring gene defects on a genome-wide scale. Multiple sample groupings and rank lists were generated to facilitate comparisons of the transcriptional responses between different PID patients and control subjects. The GXB tool enables browsing of a single transcript across studies, thereby providing new perspectives on the role of a given molecule across biological systems and PID patients. This dataset collection is available at http://pid.gxbsidra.org/dm3/geneBrowser/list.

A curated collection of transcriptome datasets to investigate the molecular mechanisms of immunoglobulin E-mediated atopic diseases

Prevalence of allergies has reached ~20% of population in developed countries and sensitization rate to one or more allergens among school age children are approaching 50%. However, the combination of the complexity of atopic allergy susceptibility/development and environmental factors has made identification of gene biomarkers challenging. The amount of publicly accessible transcriptomic data presents an unprecedented opportunity for mechanistic discoveries and validation of complex disease signatures across studies. However, this necessitates structured methodologies and visual tools for the interpretation of results. Here, we present a curated collection of transcriptomic datasets relevant to immunoglobin E-mediated atopic diseases (ranging from allergies to primary immunodeficiencies). Thirty-three datasets from the Gene Expression Omnibus, encompassing 1860 transcriptome profiles, were made available on the Gene Expression Browser (GXB), an online and open-source web application that allows for the query, visualization and annotation of metadata. The thematic compositions, disease categories, sample number and platforms of the collection are described. Ranked gene lists and sample grouping are used to facilitate data visualization/interpretation and are available online via GXB (http://ige.gxbsidra.org/dm3/geneBrowser/list). Dataset validation using associated publications showed good concordance in GXB gene expression trend and fold-change.

Using 'collective omics data' for biomedical research training

Systems-scale molecular profiling data accumulating in public repositories may constitute a useful resource for immunologists. It is for instance likely that information relevant to their chosen line of research be found among the more than 90,000 data series available in the NCBI Gene Expression Omnibus. Such 'collective omics data' may also be employed as source material for training purposes. This is the case when training curricula aim at the development of bioinformatics skills necessary for the analysis, interpretation or visualization of data generated on global scales. But 'collective omics data' may also be reused for training purposes to foster the development of the skills and 'mental habits' underpinning traditional reductionist science approaches. This review describes a small-scale initiative involving investigators, for the most part immunologists, having engaged in a range of training activities relying on 'collective omics data'.