Graph-based fuzzy clustering with validation for molecular tissue-profiling

• Recent advances in single-cell sequencing technologies allows quantification of the abundance of gene expressions in cells. This holds great promise of allowing greater understanding of functioning of the cell at the molecular level including the origin of diseases such as Alzheimer's disease, Parkinson's disease and cancer. An important part of analysing single-cell data is clustering the cells into cell types based on their corresponding gene expression profiles. In this work we identify crucial limitations of the existing Louvain clustering method. In particular, the Louvain method is shown to tend to be unable to detect rare cell types and artificially split up larger cell types into false clusters of cells. Previous cluster analysis works often lack validation of the clustering, making such artifacts difficult to detect. Moreover, the biological variation within cell types is high and cell types are often shown to continuously differentiate into one another. To this end we propose an alternative fuzzy clustering method that is shown to better cluster data based on quantitative validation indices. In addition, we present a known mathematical model of mRNA transcription that allows a more precise understanding of the underlying process generating the single-cell data. Using perturbation analysis we derive approximations of the stationary distribution of this process under the assumption of transcriptional bursts.

Ämnesord

NATURVETENSKAP  -- Matematik -- Sannolikhetsteori och statistik (hsv//swe)

NATURAL SCIENCES  -- Mathematics -- Probability Theory and Statistics (hsv//eng)

Publikations- och innehållstyp

vet (ämneskategori)

ovr (ämneskategori)