Unravelling the Biomolecular and Community Grammars of RNA Granule via Machine Learning

Non-membrane organelles, recognized as biomolecular condensates, play a vital role in cellular compartmentation, maintaining stability, and executing cellular functions. RNA granules, specific biomolecular condensates enriched in RNAs and RNA binding proteins (RBPs), are pivotal in posttranscriptional regulation and are implicated in various human diseases. The dynamic and heterogeneous nature of RNA granule compositions poses challenges in understanding their formation and functionality. Despite advancements, a comprehensive exploration of RNA granule biomolecular (i.e., protein physico-chemical properties) and community grammars (i.e., protein-protein interactions) governing their formation and functionality remains challenging. In this study, we addressed these challenges by developing two machine learning models capable of accurately identifying RNA granule (i.e., processing body (PB) and stress granule (SG)) proteome from the overall human proteome. We developed two models that incorporate protein features and achieved high accuracy, surpassing traditional liquid-liquid phase separation (LLPS) models. Our classifiers accurately distinguished RNA granule proteins from the human proteome, demonstrating their reliability in proteome-wide identification. Intriguingly, the predicted RNA granule proteome reveals a significant enrichment in biological functions associated with RNA granule-related processes, mirroring findings from established high-confidence RNA granule protein datasets. This alignment underscores the potential of our approach to construct a comprehensive RNA granule proteome. The intricate nature of biosystems cannot be comprehended merely by examining their individual components. Recognizing the complexity of biosystems, we explored the roles of protein interactions in RNA granule formation and functionality. Our analysis revealed protein-protein interaction community grammars within the RNA granule proteome, emphasizing the importance of interactions in RNA granule formations and functionality. Key clusters with dense interactions were identified, showing commonality among diverse RNA granules and involvement in crucial biological functions, such as translation, mRNA decay, rRNA processing, and mRNA splicing. This analysis proposes a hypothesis: dense protein-protein interaction clusters serve as integral functional subunits, constituting relatively stable ‘cores’ within diverse RNA granules. In conclusion, combining machine learning with the RNA granule proteome protein-protein interaction network has allowed us to explore RNA granule protein compositions and functionality in a comprehensive way. This dual approach reveals the physico-chemical properties of individual proteins (i.e., hydrophobicity of each protein) and their collective roles (i.e., highly central proteins in predicted RNA granule proteome protein-protein interaction community) in cellular processes. The findings highlight the power of using machine learning and community analysis in molecular biology to understand complex non-membrane organelles, especially RNA granules, enhancing our knowledge of cell organization and regulation.