Inferring cellular dynamics from gene expression profiles: state of the art and perspectives

Many biological phenomena rely on cellular dynamic processes whose characterization can be challenging using conventional cell culture and RNA profiling techniques. Indeed, traditional bulk protocols require the sorting of cellular subpopulations at diverse functional points to analyse the ongoing dynamics. Alternatively, cultured cells can be synchronized, a process that however is usually transient and may not allow to explore long-term phenomena. The challenges that these limitations pose, considering biological processes’ time span ranging from minutes to years, explain the urge to find new analytical procedures able to infer cellular dynamics from static data. Computationally driven approaches on single-cell data can partially overcome these problems allowing developmental studies in asynchronous cellular populations. In this direction, several research groups showed interest in designing algorithms that, starting from omics data, lead to the characterization of intermediate stages of cellular dynamics and their underlying molecular basis. Methods of trajectory inference (TI), also called pseudo-time analyses, rely on similarities in expression patterns to order cells along trajectories, with either predicted or entirely deduced topologies. Because transcriptomics is a well-developed analysis field, most of the emerged bioinformatic tools employ similarity between cells at the transcriptome level to infer pathways of cellular progression. Originating from a similar application of transcriptomic data in the reconstruction of flows through cell states, RNA velocity is a technology which deduces cell ordering from gene expression kinetics. In particular, velocyto not only generates a map of the ongoing cellular process but also deduces the directionality of cell fates and allows extrapolation of future transcriptomic states through the ratio between pre-mRNA and mRNA copy numbers of different genes. While these techniques move forward and many successes confirm their efficacy as analytical methods, new tools have been developed to process output data from these algorithms and further dig into the molecular biology of cells. TENET, for instance, is a tool designed to predict large-scale gene regulatory cascades. The algorithm measures the strength of causality between couple of genes, considering the dynamics of genes’ expression in cells once reconducted to subsequent developmental states by TI analysis. This technique provides annotations of directionality for gene co-expression networks in a high throughput manner, thus enabling preliminary the bioinformatic analyses in research projects to more accurately predict the roles and importance of genes in a biological process. In this thesis I will introduce the main approaches adopted towards inferring cellular dynamics, showing also their relevance in the study of transcriptomes to deduce causal relationships between genes. For this purpose, I will mainly describe velocyto, as an algorithm of RNA velocity, and TENET, to measure potential causality in RNA expression.