Search

Multi-omics studies have enabled us to understand the mechanistic drivers behind complex disease states and progressions, thereby providing novel and actionable biological insights into health status. However, integrating data from multiple modalities is challenging due to the high dimensionality of data and noise associated with each platform. Non-overlapping features and technical batch effects in the data make the task of learning more complicated. Conventional machine learning (ML) tools are not quite effective against such data integration hazards. In addition, existing methods for single cell multi-omics integration are computationally expensive. This has encouraged the development of a novel architecture that produces a robust model for integration of high-dimensional multi-omics data, which would be capable of learning meaningful features for further downstream analysis. In this work, we have introduced a novel Unsupervised neural network for single cell Multi-omics INTegration (UMINT). UMINT serves as a promising model for integrating variable number of single cell omics layers with high dimensions, and provides substantial reduction in the number of parameters. It is capable of learning a latent low-dimensional embedding that can capture useful data characteristics. The effectiveness of UMINT has been evaluated on benchmark CITE-seq (paired RNA and surface proteins) datasets. It has outperformed existing state-of-the-art methods for multi-omics integration.

BackgroundDietary high fructose (HFr) is a known metabolic disruptor contributing to development of obesity and diabetes in Western societies. Initial molecular changes from exposure to HFr on liver metabolism may be essential to understand the perturbations leading to insulin resistance and abnormalities in lipid and carbohydrate metabolism. We studied vervet monkeys (Clorocebus aethiops sabaeus) fed a HFr (n=5) or chow diet (n=5) for 6 weeks, and obtained clinical measures of liver function, blood insulin, cholesterol and triglycerides. In addition, we performed untargeted global transcriptomics, proteomics, and metabolomics analyses on liver biopsies to determine the molecular impact of a HFr diet on coordinated pathways and networks that differed by diet.ResultsWe show that integration of omics data sets improved statistical significance for some pathways and networks, and decreased significance for others, suggesting that multiple omics datasets enhance confidence in relevant pathway and network identification. Specifically, we found that sirtuin signaling and a peroxisome proliferator activated receptor alpha (PPARA) regulatory network were significantly altered in hepatic response to HFr. Integration of metabolomics and miRNAs data further strengthened our findings.ConclusionsOur integrated analysis of three types of omics data with pathway and regulatory network analysis demonstrates the usefulness of this approach for discovery of molecular networks central to a biological response. In addition, metabolites aspartic acid and docosahexaenoic acid (DHA), protein ATG3, and genesATG7, HMGCS2link sirtuin signaling and the PPARA network suggesting molecular mechanisms for altered hepatic gluconeogenesis from consumption of a HFr diet.

High grade serous ovarian cancer (HGSOC) is a major unmet need in oncology. The persistent uncertainty on its originating tissue has contributed to hamper the discovery of oncogenic pathways and effective therapies. Here we define the DNA methylation print that distinguishes the human fimbrial (FI) and ovarian surface epithelia (OSE) and develop a robust epigenetic cell-of-origin tracer that stratifies HGSOC in FI-and OSE-originated tumors across all available cohorts. We translate this origin-based stratification into a clinically actionable transcriptomic signature, demonstrating its prognostic impact on patients’ survival and identifying novel network level dysregulations specific for the two disease subtypes.