ESAE-SDA: ensemble sparse autoencoder framework for epigenomics-informed snoRNA-disease associations prediction.

Area of Science:

• Genomics and Epigenomics
• Non-coding RNA Biology
• Computational Biology

Background:

• Small nucleolar RNAs (snoRNAs) are crucial non-coding RNAs involved in RNA modifications and epigenetic regulation.
• Identifying snoRNA-disease associations (SDAs) is vital for understanding epigenomic dysregulation in diseases.
• Existing AI methods for SDA prediction face challenges like sample imbalance and high false-negative rates.

Purpose of the Study:

• To develop an advanced AI model, ESAE-SDA, for accurate prediction of snoRNA-disease associations.
• To overcome limitations of existing methods, including sample imbalance and false negatives.
• To enhance the understanding of epigenomic roles in complex diseases.

Main Methods:

• Developed ESAE-SDA, integrating sparse autoencoders and ensemble learning.
• Constructed comprehensive snoRNA-disease representations using multi-source similarity.
• Employed k-means clustering for negative sample selection and deep sparse autoencoders for feature learning.
• Utilized GNN-based learners with dynamic resampling and weighted fusion for robust inference.

Main Results:

• ESAE-SDA demonstrated superior performance compared to state-of-the-art methods on a public SDA dataset.
• The model achieved enhanced robustness and generalization through its ensemble approach.
• A case study on ophthalmic diseases identified potential epigenetically relevant snoRNAs.

Conclusions:

• ESAE-SDA is a powerful tool for predicting snoRNA-disease associations, outperforming existing methods.
• The model aids in uncovering novel epigenetically relevant snoRNAs with therapeutic potential.
• ESAE-SDA contributes significantly to epigenomics-driven disease research and target discovery.