Many machine learning applications in bioinformatics currently rely on matching gene identities when analyzing input gene signatures and fail to take advantage of preexisting knowledge about gene functions. To further enable comparative analysis of OMICS datasets, including target deconvolution and mechanism of action studies, we develop an approach that represents gene signatures projected onto their biological functions, instead of their identities, similar to how the word2vec technique works in natural language processing. We develop the Functional Representation of Gene Signatures (FRoGS) approach by training a deep learning model and demonstrate that its application to the Broad Institute’s L1000 datasets results in more effective compound-target predictions than models based on gene identities alone. By integrating additional pharmacological activity data sources, FRoGS significantly increases the number of high-quality compound-target predictions relative to existing approaches, many of which are supported by in silico and/or experimental evidence. These results underscore the general utility of FRoGS in machine learning-based bioinformatics applications. Prediction networks pre-equipped with the knowledge of gene functions may help uncover new relationships among gene signatures acquired by large-scale OMICs studies on compounds, cell types, disease models, and patient cohorts.

Many viral pathogens target innate sensing cascades and/or cellular transcription factors to suppress antiviral immune responses. Here, we show that the accessory viral protein U (Vpu) of HIV-1 exerts broad immunosuppressive effects by inhibiting activation of the transcription factor NF-κB. Global transcriptional profiling of infected CD4 +T cells revealed that vpu-deficient HIV-1 strains induce substantially stronger immune responses than the respective wild type viruses. Gene set enrichment analyses and cytokine arrays showed that Vpu suppresses the expression of NF-κB targets including interferons and restriction factors. Mutational analyses demonstrated that this immunosuppressive activity of Vpu is independent of its ability to counteract the restriction factor and innate sensor tetherin. However, Vpu-mediated inhibition of immune activation required an arginine residue in the cytoplasmic domain that is critical for blocking NF-κB signaling downstream of tetherin. In summary, our findings demonstrate that HIV-1 Vpu potently suppresses NF-κB-elicited antiviral immune responses at the transcriptional level.

Influenza A Virus (IAV) is a recurring respiratory virus with limited availability of antiviral therapies. Understanding host proteins essential for IAV infection can identify targets for alternative host-directed therapies (HDTs). Using affinity purification-mass spectrometry and global phosphoproteomic and protein abundance analyses using three IAV strains (pH1N1, H3N2, H5N1) in three human cell types (A549, NHBE, THP-1), we map 332 IAV-human protein-protein interactions and identify 13 IAV-modulated kinases. Whole exome sequencing of patients who experienced severe influenza reveals several genes, including scaffold protein AHNAK, with predicted loss-of-function variants that are also identified in our proteomic analyses. Of our identified host factors, 54 significantly alter IAV infection upon siRNA knockdown, and two factors, AHNAK and coatomer subunit COPB1, are also essential for productive infection by SARS-CoV-2. Finally, 16 compounds targeting our identified host factors suppress IAV replication, with two targeting CDK2 and FLT3 showing pan-antiviral activity across influenza and coronavirus families. This study provides a comprehensive network model of IAV infection in human cells, identifying functional host targets for pan-viral HDT.

AIR data sharing is integral to spur research reproducibility, promote data reuse, and accelerate research. However, the first step in using these data assets-discovering that they exist-is compounded by problems in incentives, standardization, and coordination of efforts. In 2023, the National Institutes of Health (NIH) implemented their updated Data Sharing Policy, which mandates timely data sharing of all NIH-funded data. However, for the policy to shift the data sharing culture, to improve research reproducibility, and to promote data reuse, several changes must happen. In a companion piece 1 , we present a survey of the data sharing landscape for immune-mediated and infectious disease data, combined with our efforts to create a reusable methodology to catalog data we generate. We found that researchers routinely share data, but datasets are still not necessarily findable, accessible, interoperable, or reusable (FAIR) 2. In the course of these efforts, we identified three interdependent barriers that need to be addressed to maximize the impact of data sharing efforts that are becoming increasingly commonplace: (1) a lack of incentives, (2) little standardization in metadata collection and access, and (3) uncoordinated efforts. Here, we propose solutions to improve the FAIRness of metadata, data, and the experimental conditions used to generate these data. Our solutions include leveraging journals as a key driver of research incentives and consolidating data searching and citation into a centralized platform. While we focus our discussion based on our experience primarily with NIH-funded research, these principles are broadly applicable to other data sharing efforts worldwide. Barrier 1: Without incentives, researchers tend to provide incomplete metadata, which limits data discovery and reuse. Currently, there are few tangible incentives for researchers to share data. There is limited recognition amongst scientists regarding the benefits of data sharing, while the potential for downside effects such as the risk of getting scooped, concerns over protecting participant/patient privacy, a lack of Intellectual Property protection, and the enhanced workload associated with providing good, reusable, and discoverable data are all well recognized. As a result, researchers frequently do the bare minimum to comply with journal and funder mandates. When sharing data, researchers also include metadata which describe the contents of the dataset to aid in searchability, but these metadata are often insufficient to help researchers identify useful datasets. Even if a metadata property is required, researchers can circumvent these requirements by providing short, generic descriptions. Consequently, metadata is often cursory and incomplete, and data is often provided in non-machine readable formats (e.g., .pdf, Microsoft Word, or free text) without associated documentation to support reuse. While there are many projects that attempt to improve the FAIRness of poorly collected metadata 3,4 , it is far easier to collect complete, well-structured metadata from the start (see Solution 2.3). Our proposed solutions tap into the two primary incentive structures for researchers: funding opportunities and recognition via citations. Solution 1.1: Ensure that there are sufficient resources for data sharing activities within funding opportunities. Documenting data to make it reusable takes time and effort, detracting from experimentation, data analysis, and synthesis of insights into publications. If funding agencies (e.g., NIH) mandate data sharing, they should be prepared to increase budgets accordingly to dedicate a portion of the effort to ensuring the resulting CommeNt