Abstract: Modeling genes regulation networks is a major yet challenging stake to understand physiopathology. We show that genes belonging to the same regulation network have common geometrical variations when their biological function is modified by an environmental condition. This allows the first topological approach to transcriptomes analysis. First we pre-process the genes expressions data set with quantile normalization, logarithmization, removing the low expressed genes, and finally replacing the values by their z-scores. Then we use UMAP, a dimensional reduction algorithm topology preserving, to sum up the conditions observed. The local regulation networks with common behavior tend to set apart. We use a db-scan to isolate those clusters. Finally, we estimate the investment likeliness
for each cluster by measuring their individual prediction performance on the studied condition. Our method provides strong inferences on which genes are implied in the cell reaction. Moreover, our method provides leads for common transcription factors among genes concerned in specific pathological situations, and thus for new therapeutic targets.

Keywords: Genes regulation network ; Machine learning; Omics data; UMAP

Abu-Jamous (2018). Clust: automatic extraction of optimal co-expressed gene clusters from gene expression data. Genome Biology.
Camara (2017). Topological methods for genomics: present and future directions. Curr Opin SystBiol.
Dorrity (2020). Dimensionality reduction by UMAP to visualize physical and genetic interactions. Nature Communications.
Luo (2021). A topology-preserving dimensionality reduction method for single-cell RNA-seq data using graph autoencoder. Scientific Reports.