Deep learning models for modeling cellular transcription systems

摘要: Cellular signal transduction system (CSTS) plays a fundamental role in maintaining homeostasis of cell by detecting changes its environment and orchestrates response. Perturbations CSTS lead to diseases such as cancers. Almost all CSTSs are involved regulating the expression certain genes leading signature gene expression. Therefore, profile is readout state could be used infer CSTS. However, convoluted mixture responses active signaling pathways cells. Therefore it difficult find associated with an individual pathway. An efficient way de-convoluting signals embedded needed. At beginning thesis, we applied Pearson correlation coefficient analysis study cellular transduced from ceramide species (lipids) genes. We found significant correlations between specific or groups showed that various dihydroceramide families regulated distinct subsets target predicted participate biologic processes. it’s well known pathway structure hierarchical. Useful information may not fully detected if only linear models More complex non-linear needed represent hierarchical This motivated us investigate contemporary deep learning (DLMs). Later, learn distributed representation statistical structures transcriptomic data. The organization machinery. Besides, they provide abstract data flexibility different degrees granularity. were capable biologically sensible representations (e.g., hidden units first layer transcription factors) revealing novel insights regarding machinery also model outperformed state-of-the-art methods Elastic-Net Linear Regression, Support Vector Machine Non-Negative Matrix Factorization.