Discovering Novel Ciliary Genes Through Machine Learning

Abstract

The primary cilium is an evolutionarily conserved organelle present in many eukaryotic cells, where it plays a central role in cellular signaling and environmental sensing (Hilgendorf et al. 2024). Despite its importance, the complete set of genes required for its structure and function remains incompletely characterized. Although cilia have a characteristic protein composition, no consensus sequences or domains enable their systematic identification. We hypothesized that ciliary genes share expression-based and other biological features, independent of DNA sequence similarity, that can be used for their identification. Caenorhabditis elegans provides a powerful model for studying ciliary biology due to its well-characterized nervous system, ease of experimental manipulation, and abundant publicly available transcriptomic data. We constructed a transcriptomic dataset from publicly available single-cell RNA sequencing data (Packer et al. 2019; Taylor et al. 2021). Agglomerative clustering was performed independently on embryo and adult datasets, and clusters enriched in ciliary genes were identified. Genes co-clustering with known ciliary genes were selected as candidates for further analysis. Functional Landscape Arrays (FLAs) provide a numerical representation of a gene's genomic context (Pazos Obregón et al. 2022). Weighted FLAs (wFLAs) were constructed using genomic windows of different sizes around each gene and quantifying the presence of functional groups within these regions. These features were used to train and evaluate different classifiers, and the best-performing model was applied genome-wide to generate a second set of candidate genes. Finally, candidates from both approaches were integrated, yielding a preliminary set of 56 genes potentially associated with ciliary function.