Inferência de redes de regulação gênica a partir de séries temporais via meta-heurísticas

Abstract

The inference of gene regulatory networks (GRNs) is a difficult and important problem, with challenges largely addressed in the area called Systems Biology. Its applications include biotechnology and health, assisting in the development of drugs, since understanding patterns in gene interactions can lead to important discoveries related to diseases in organisms. Single-cell RNA sequencing (scRNA-Seq) has provided unprecedented resolution to the field of transcriptomics. Experiments using scRNA-Seq are attractive for the inference of GRNs due to the generation of thousands of independent measurements and the possibility of obtaining a more accurate pseudotemporal view of the dynamics of gene expression. However, not all genes are expressed all the time. The selection of gene subsets that model the desired biological phenomenon also constitutes a challenge for the inference of GRNs. Boolean networks and those modeled through systems of ordinary differential equations (ODEs) are commonly used to represent GRNs. Nevertheless, there is no standard method for discretizing the data that is provided to Boolean networks. Boolean networks can be modeled in the form of digital circuits. Among evolutionary computing techniques, Cartesian Genetic Programming (CGP) is considered the most efficient technique for the evolution and optimization of combinational logic circuits. However, evolutionary computing techniques do not appear among the algorithms highlighted as state of the art for reconstructing GRNs, mainly motivated by scalability problems. Furthermore, the lack of knowledge about ground-truth networks and the non-standardization of the way to attribute quality to an inferred network increase the challenge when solving the problem. In this work, we propose a framework that uses CGP to infer Boolean GRNs and obtain a continuous model from data in the form of time series. Each step of the proposed framework is explored, covering the pre-processing of gene expression data, the selection of subsets of genes via clustering techniques as a way of directing the search process, the ways in which the data should be discretized in order to obtain a Boolean model, the behavior of the genetic variation operators in the CGP, the way in which a Boolean model can be converted into a system of ODEs and the determination of the numerical coefficients of this system of ODEs via Evolutionary Strategies. A new procedure for discretizing gene expression data in the form of time series is also proposed. Finally, a review of the methodological process adopted in the context of inferring gene regulation networks from scRNA-Seq data, covering the intrinsic characteristics of sequencing technology, the selection of genes of interest, the network motifs , the reference networks and the metrics and way to evaluate the inferred networks are presented. As a result, a new methodological process is proposed. All proposals are evaluated in benchmark problems, which consider synthetic and real data obtained through microarrays and scRNA-Seq, data from stochastic simulation, in addition to data from organisms widely known and explored in the literature , such as Saccharomyces cerevisiae and Escherichia coli, and data from the DREAM4 competition. The results show that the proposals are superior or competitive with state-of-the-art methods for the inference of GRNs and provide an interpretable solution that can assist domain experts in the field of Systemic Biology. Furthermore, the proposed methodological process makes the comparison of different GRN inference algorithms fairer.