The rapid advancement of high-throughput sequencing and other assay technologies has resulted in the generation of large and complex multi-omics datasets, offering unprecedented opportunities for advancing precision medicine. However, multi-omics data integration remains challenging due to the high-dimensionality, heterogeneity, and frequency of missing values across data types. Computational methods leveraging statistical and machine learning approaches have been developed to address these issues and uncover complex biological patterns, improving our understanding of disease mechanisms. Here, we comprehensively review state-of-the-art multi-omics integration methods with a focus on deep generative models, particularly variational autoencoders (VAEs) that have been widely used for data imputation, augmentation, and batch effect correction. We explore the technical aspects of VAE loss functions and regularisation techniques, including adversarial training, disentanglement, and contrastive learning. Moreover, we highlight recent advancements in foundation models and multimodal data integration, outlining future directions in precision medicine research.

MOTIVATION: Phosphoproteomic experiments are increasingly used to study the changes in signaling occurring across different conditions. It has been proposed that changes in phosphorylation of kinase target sites can be used to infer when a kinase activity is under regulation. However, these approaches have not yet been benchmarked due to a lack of appropriate benchmarking strategies. RESULTS: We used curated phosphoproteomic experiments and a gold standard dataset containing a total of 184 kinase-condition pairs where regulation is expected to occur to benchmark and compare different kinase activity inference strategies: Z-test, Kolmogorov Smirnov test, Wilcoxon rank sum test, gene set enrichment analysis (GSEA), and a multiple linear regression model. We also tested weighted variants of the Z-test and GSEA that include information on kinase sequence specificity as proxy for affinity. Finally, we tested how the number of known substrates and the type of evidence ( in vivo , in vitro or in silico ) supporting these influence the predictions. CONCLUSIONS: Most models performed well with the Z-test and the GSEA performing best as determined by the area under the ROC curve (Mean AUC = 0.722). Weighting kinase targets by the kinase target sequence preference improves the results marginally. However, the number of known substrates and the evidence supporting the interactions has a strong effect on the predictions. AVAILABILITY AND IMPLEMENTATION: The KSEA implementation is available in https://github.com/ evocellnet/ksea. Additional data is available in http://phosfate.com. CONTACT: [email protected] or [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Cells react to extracellular perturbations with complex and intertwined responses. Systematic identification of the regulatory mechanisms that control these responses is still a challenge and requires tailored analyses integrating different types of molecular data. Here we acquired time-resolved metabolomics measurements in yeast under salt and pheromone stimulation and developed a machine learning approach to explore regulatory associations between metabolism and signal transduction. Existing phosphoproteomics measurements under the same conditions and kinase-substrate regulatory interactions were used to in silico estimate the enzymatic activity of signalling kinases. Our approach identified informative associations between kinases and metabolic enzymes capable of predicting metabolic changes. We extended our analysis to two studies containing transcriptomics, phosphoproteomics and metabolomics measurements across a comprehensive panel of kinases/phosphatases knockouts and time-resolved perturbations to the nitrogen metabolism. Changes in activity of transcription factors, kinases and phosphatases were estimated in silico and these were capable of building predictive models to infer the metabolic adaptations of previously unseen conditions across different dynamic experiments. Time-resolved experiments were significantly more informative than genetic perturbations to infer metabolic adaptation. This difference may be due to the indirect nature of the associations and of general cellular states that can hinder the identification of causal relationships. This work provides a novel genome-scale integrative analysis to propose putative transcriptional and post-translational regulatory mechanisms of metabolic processes.

SUMMARY: Drug versus Disease (DvD) provides a pipeline, available through R or Cytoscape, for the comparison of drug and disease gene expression profiles from public microarray repositories. Negatively correlated profiles can be used to generate hypotheses of drug-repurposing, whereas positively correlated profiles may be used to infer side effects of drugs. DvD allows users to compare drug and disease signatures with dynamic access to databases Array Express, Gene Expression Omnibus and data from the Connectivity Map. AVAILABILITY AND IMPLEMENTATION: R package (submitted to Bioconductor) under GPL 3 and Cytoscape plug-in freely available for download at www.ebi.ac.uk/saezrodriguez/DVD/.

BACKGROUND: A standard graphical notation is essential to facilitate exchange of network representations of biological processes. Towards this end, the Systems Biology Graphical Notation (SBGN) has been proposed, and it is already supported by a number of tools. However, support for SBGN in Cytoscape, one of the most widely used platforms in biology to visualise and analyse networks, is limited, and in particular it is not possible to import SBGN diagrams. RESULTS: We have developed CySBGN, a Cytoscape plug-in that extends the use of Cytoscape visualisation and analysis features to SBGN maps. CySBGN adds support for Cytoscape users to visualize any of the three complementary SBGN languages: Process Description, Entity Relationship, and Activity Flow. The interoperability with other tools (CySBML plug-in and Systems Biology Format Converter) was also established allowing an automated generation of SBGN diagrams based on previously imported SBML models. The plug-in was tested using a suite of 53 different test cases that covers almost all possible entities, shapes, and connections. A rendering comparison with other tools that support SBGN was performed. To illustrate the interoperability with other Cytoscape functionalities, we present two analysis examples, shortest path calculation, and motif identification in a metabolic network. CONCLUSIONS: CySBGN imports, modifies and analyzes SBGN diagrams in Cytoscape, and thus allows the application of the large palette of tools and plug-ins in this platform to networks and pathways in SBGN format.