Orthrus: A New AI Model Advancing RNA Genetics and Gene Regulation Prediction

    A recent publication introduces Orthrus, a cutting edge RNA foundation model designed to improve how scientists predict RNA behavior and gene regulation. Despite the massive amount of genomic data available today, understanding the “RNA regulatory code” remains a major challenge in genetics. Traditional experimental methods like eCLIP and ribosome profiling are accurate but expensive and time-consuming, creating a need for faster computational alternatives.

    Orthrus addresses this gap by using a machine learning approach called contrastive learning, combined with a Mamba-based encoder optimized for long RNA sequences. Unlike older models that rely on generic text based training methods, Orthrus is trained using biologically meaningful relationships, specifically RNA splice variants and evolutionary similarities across species. The model learns by comparing related RNA sequences from over 400 mammalian species, allowing it to better capture functional genetic relationships.

    This approach significantly improves performance in predicting key RNA properties such as RNA half-life, ribosome load, protein localization, and gene function classification. Importantly, Orthrus performs well even in low-data environments, reducing the need for large labeled datasets, which is a major limitation in genetics research.

    Overall, Orthrus represents a shift toward more biologically informed artificial intelligence models in genomics. By integrating evolutionary biology with machine learning, it improves our ability to interpret RNA function and gene regulation more accurately than previous self-supervised models.

Article link: https://www.marktechpost.com/2024/10/15/orthrus-a-contrastive-learning-approach-for-enhanced-rna-representation-and-property-prediction/

Additional resource: https://www.biorxiv.org/content/10.1101/2024.10.10.617658v1.full.pdf