Unsupervised neural network for single cell Multi-omics INTegration (UMINT): an application to health and disease.

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 high dimensionality and diverse nature of data, and noise associated with each platform. Sparsity in data, non-overlapping features and technical batch effects make the task of learning more complicated. Conventional machine learning (ML) tools are not quite effective against such data integration hazards due to their simplistic nature with less capacity. In addition, existing methods for single cell multi-omics integration are computationally expensive. Therefore, 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. It has a light-weight architecture with substantially reduced number of parameters. The proposed model is capable of learning a latent low-dimensional embedding that can extract useful features from the data facilitating further downstream analyses. UMINT has been applied to integrate healthy and disease CITE-seq (paired RNA and surface proteins) datasets including a rare disease Mucosa-Associated Lymphoid Tissue (MALT) tumor. It has been benchmarked against existing state-of-the-art methods for single cell multi-omics integration. Furthermore, UMINT is capable of integrating paired single cell gene expression and ATAC-seq (Transposase-Accessible Chromatin) assays as well.

EZH2-Mediated H3K27me3 Targets Transcriptional Circuits of Neuronal Differentiation.

The Polycomb Repressive Complex 2 (PRC2) plays important roles in the epigenetic regulation of cellular development and differentiation through H3K27me3-dependent transcriptional repression. Aberrant PRC2 activity has been associated with cancer and neurodevelopmental disorders, particularly with respect to the malfunction of sits catalytic subunit EZH2. Here, we investigated the role of the EZH2-mediated H3K27me3 apposition in neuronal differentiation. We made use of a transgenic mouse model harboring Ezh2 conditional KO alleles to derive embryonic stem cells and differentiate them into glutamatergic neurons. Time course transcriptomics and epigenomic analyses of H3K27me3 in absence of EZH2 revealed a significant dysregulation of molecular networks affecting the glutamatergic differentiation trajectory that resulted in: (i) the deregulation of transcriptional circuitries related to neuronal differentiation and synaptic plasticity, in particular LTD, as a direct effect of EZH2 loss and (ii) the appearance of a GABAergic gene expression signature during glutamatergic neuron differentiation. These results expand the knowledge about the molecular pathways targeted by Polycomb during glutamatergic neuron differentiation.

A cell-of-origin epigenetic tracer reveals clinically distinct subtypes of high-grade serous ovarian cancer.

BACKGROUND: High-grade serous ovarian cancer (HGSOC) is a major unmet need in oncology. The remaining uncertainty on its originating tissue has hampered the discovery of molecular oncogenic pathways and the development of effective therapies. METHODS: We used an approach based on the retention in tumors of a DNA methylation trace (OriPrint) that distinguishes the two putative tissues of origin of HGSOC, the fimbrial (FI) and ovarian surface epithelia (OSE), to stratify HGSOC by several clustering methods, both linear and non-linear. The identified tumor subtypes (FI-like and OSE-like HGSOC) were investigated at the RNAseq level to stratify an in-house cohort of macrodissected HGSOC FFPE samples to derive overall and disease-free survival and identify specific transcriptional alterations of the two tumor subtypes, both by classical differential expression and weighted correlation network analysis. We translated our strategy to published datasets and verified the co-occurrence of previously described molecular classification of HGSOC. We performed cytokine analysis coupled to immune phenotyping to verify alterations in the immune compartment associated with HGSOC. We identified genes that are both differentially expressed and methylated in the two tumor subtypes, concentrating on PAX8 as a bona fide marker of FI-like HGSOC. RESULTS: We show that: - OriPrint is a robust DNA methylation tracer that exposes the tissue of origin of HGSOC. - The tissue of origin of HGSOC is the main determinant of DNA methylation variance in HGSOC. - The tissue of origin is a prognostic factor for HGSOC patients. - FI-like and OSE-like HGSOC are endowed with specific transcriptional alterations that impact patients' prognosis. - OSE-like tumors present a more invasive and immunomodulatory phenotype, compatible with its worse prognostic impact. - Among genes that are differentially expressed and regulated in FI-like and OSE-like HGSOC, PAX8 is a bona fide marker of FI-like tumors. CONCLUSIONS: Through an integrated approach, our work demonstrates that both FI and OSE are possible origins for human HGSOC, whose derived subtypes are both molecularly and clinically distinct. These results will help define a new roadmap towards rational, subtype-specific therapeutic inroads and improved patients' care.

Estimating gene expression from DNA methylation and copy number variation: A deep learning regression model for multi-omics integration.

Gene expression analysis plays a significant role for providing molecular insights in cancer. Various genetic and epigenetic factors (being dealt under multi-omics) affect gene expression giving rise to cancer phenotypes. A recent growth in understanding of multi-omics seems to provide a resource for integration in interdisciplinary biology since they altogether can draw the comprehensive picture of an organism's developmental and disease biology in cancers. Such large scale multi-omics data can be obtained from public consortium like The Cancer Genome Atlas (TCGA) and several other platforms. Integrating these multi-omics data from varied platforms is still challenging due to high noise and sensitivity of the platforms used. Currently, a robust integrative predictive model to estimate gene expression from these genetic and epigenetic data is lacking. In this study, we have developed a deep learning-based predictive model using Deep Denoising Auto-encoder (DDAE) and Multi-layer Perceptron (MLP) that can quantitatively capture how genetic and epigenetic alterations correlate with directionality of gene expression for liver hepatocellular carcinoma (LIHC). The DDAE used in the study has been trained to extract significant features from the input omics data to estimate the gene expression. These features have then been used for back-propagation learning by the multilayer perceptron for the task of regression and classification. We have benchmarked the proposed model against state-of-the-art regression models. Finally, the deep learning-based integration model has been evaluated for its disease classification capability, where an accuracy of 95.1% has been obtained.