Análise de biclusterização em células únicas

Abstract

The single-cell analysis has gained prominence in the last few years, propelled by technological advances that made RNA sequencing more accessible and reduced information loss. Initially, due to technical limitations, sequencing was performed in bulk: RNA was extracted from a sample composed of multiple cells, resulting in an average gene expression profile. This method allowed comparisons between healthy and diseased tissues, or between different experimental conditions, based on average gene expression. However, this approach didn’t capture the cellular heterogeneity within the same tissue. With the advent of single-cell RNA sequencing, it became possible to analyze each cell of the sample individually, enabling the identification of cell subpopulations and characterization of subtle variations among cells from the same region, something specially relevant in tissues with high cellular diversity, like the brain. Despite its advantages, single-cells data presents specific challenges, such as high levels of noise, a large proportion of zeros (sparsity) and high dimensionality, since each cell is described by thousands of genes. In this study, normalization techniques were explored to process the data and reduce the impact of technical effects, as well as biclustering methods to simultaneously group cell with similar profiles and the most expressive genes in each group. To this end, a gene expression dataset composed of 18,928 genes and 734 cells derived from the human fetal brain and cerebral organoids obtained at different developmental stages was analyzed. The normalization was able to reduce the effects of technical variation by detecting relevant genes with low expression levels, while biclustering enabled the simultaneous identification of groups of cells with similar expression profiles and the most representative genes within each group. The results indicated that more mature organoids exhibit greater similarity to fetal cells, although they still do not fully reproduce all the characteristics of the human fetal brain.