Single-cell genomics, rooted in single-cell sequencing, has great potential for providing insight into fundamental questions in biomedical science and drive new health science discoveries, such as: How many cell types and functional states are there in a given tissue? What is the range of natural variation within a cell type and how is such variability affected by genetic and environmental factors? What happens at the single-cell level during cell fate determination in the developmental process? How does cellular heterogeneity with a tumor affect response to therapy and how can we address this with precision medicine? The list is endless. However, the explosive growth of single-cell sequencing technologies also brings new computational challenges. One major challenge is the “sparse read counts data”: because of the minuscule amount of genetic material in a single cell, fragments of the genome are often missing from the sequencing read-out, and existing tools are ineffective in addressing this missing-data problem and piecing together reliable genomic information.
The research team will establish a Michigan Center for Single-Cell Genomic Data Analytics, and connect mathematicians and data scientists with biological researchers to develop, evaluate, and implement a variety of cutting-edge methodologies in sparse data analysis. These methodologies will address issues in data normalization, batch effect detection and correction, marker selection, classification, rare class identification, differential expression, network and phylogenetic inference, develop tools for cyclic or time-series data, and enable information integration across data types. The team will apply these methodologies to four biological questions to test their utility: 1) Intra-tumor heterogeneity, cancer stem cells in metastasis and treatment resistance, cancer genome evolution; 2) spermatogenesis as a model for cell fate determination during development; 3) transcriptional complexity and gene regulation at the single-cell level; 4) molecular changes at the single-cell level as a result of environmental exposures and windows of susceptibility.
The outcome from this research project will have much broader impact on biomedical research beyond the four research areas that will be used as test cases. Sparse data analytics also has wide application beyond health sciences. For example, Electronic Health Records are inherently sparse, as are consumer data (purchasing, rating, or video viewing habits), location and usage data of mobile devices, connectivity in social networks, medical imaging or land imaging by satellites. In short, this line of research is conceptually connected with many areas of active research in data science and will produce general-purpose tools for many research areas.