Zeballos - Goron, S., Salinas, G., Pazos Obregon, F.
This paper presents a method to discover new oxidative phosphorylation (OXPHOS) genes in the model organism C. elegans by combining transcriptomic resolution and machine learning strategies.
Gene function annotation remains a major challenge, and understanding of the OXPHOS process in non-mammalian lineages is limited. Existing homology-based approaches may overlook key genes lacking detectable sequence similarity.
Using a curated set of 65 known OXPHOS genes, the authors trained an ensemble of supervised learning models on time-resolved bulk RNA-seq transcriptomes of C. elegans. To address annotation uncertainty, they implemented a novel informed bagging strategy with a two-round training scheme, where weak positives were initially excluded and later incorporated based on model predictions. In parallel, they performed cluster-based functional inference using embryonic and adult single-cell RNA-seq datasets.
The integration of both approaches produced a list of novel candidate genes with high likelihood of OXPHOS function. This work provides a powerful framework for predicting gene function using transcriptomic patterns without relying on sequence homology, with translational relevance for studying OXPHOS as a therapeutic target in parasitic helminths.