Abstract: The real-world clinical bulk sequencing samples contain mosaics of various cell types. The recent success of signal deconvolution methods demonstrated that properly accounting for cell type mixture proportions leads to improvement in biomarker discovery and better interpretations at cell-type resolution. The current statistical methods, however, are all based on a very strong assumption that a common reference panel is shared across the whole population. This deviates from the biological fact that person-to-person heterogeneity exists, even at cell-type level. To address these issues, I propose to develop a series of novel statistical methods to conduct deconvolution at the `personalized + cell-type' level. I will properly model the admixture data, retrieve individual- specific and cell-type-specific reference panels, use them to improve cell type proportion deconvolution, and conduct statistical rigorous test to identify cell-type-specific differentially-expression genes. First, leveraging on temporal or repeatedly measured data, we will develop a mixed-effect model based iterative algorithm, to extend the current cell type mixture modeling by allowing for personal-level reference panels. We will obtain individual- specific and cell-type-specific reference estimation, to retrieve more accurate, personalized gene expression profiles. Next, we will develop a statistical framework to deconvolute samples with repeated measures, to im- prove the cell type proportion estimation. We will also conduct wet-lab experiments to validate our statistical methods. Third, we will develop methods to test for cell-type-specific differentially expressed genes, by incorpo- rating individual-specific reference panels. We will also compile existing tools that can conduct cell-type-level differential expression gene analysis in the bulk data, benchmark them, and develop methods to evaluate the statistical power. Finally, we will apply the proposed methods to analyze the bulk transcriptome data from The Environmental Determinants of Diabetes in the Young (TEDDY) for type 1 diabetes research, from Parkinson's Disease Biomarkers Program (PDBP), and other large consortia with longitudinal bulk samples. Our methods and software packages will provide important resources that enable new biomedical genomics studies, such as biomarker discovery of individual cell-type transcriptomics/epigenetics biomarkers associated with environmen- tal factors, or disease risk prediction using cell type related profiles and proportions. Our proposed work will significantly enhance our abilities to re-analyze and re-use bulk sequencing data, enhance the utility of decon- volution to a `personalized + cell-type' level, and have major impact on the cell-type-specific data mining and inference in clinical settings.