Thumbnails:
List:
Year:
Category:
Session:
Poster:
Getting poster data...
Étienne Fafard-Couture, Pierre-Étienne Jacques & Michelle S. Scott (Université de Sherbrooke, Sherbrooke, Canada)
Small nucleolar RNAs (snoRNAs) are non-coding RNAs involved in the regulation of ribosome biogenesis and splicing. Mainly located in the introns of host genes, snoRNAs are highly abundant across human tissues. However, little is known about the factors regulating their expression. For instance, only a third of all known snoRNAs are expressed in human, with snoRNAs encoded in the same host gene often differing drastically in their level. By integrating large transcriptomic datasets within a machine learning (ML) approach, we built models that accurately predict whether a snoRNA is expressed or not. Using SHAP values, we identified which features were the most relevant in the predictions. We find that snoRNA motif conservation, stability and expression of their host locus are the main snoRNA expression determinants. Applying our models to other vertebrates, we observe a conserved low proportion of expressed snoRNAs, with a significant anticorrelation between the number of snoRNA per species and the proportion of expressed snoRNAs. Overall, our approach illustrates that by interpreting ML models that were trained on high-quality datasets, we can gain valuable biological insights.