AI foundation models for RNA biology.

Area of Science:

• RNA Biology
• Artificial Intelligence
• Bioinformatics

Background:

• RNA biology is experiencing a paradigm shift driven by AI.
• Foundation models are trained on massive RNA datasets to understand sequence, structure, and function.
• Self-supervised learning enables generalizable RNA representations.

Purpose of the Study:

• To provide a comprehensive review of RNA foundation models.
• To explain the key components: datasets, architectures, self-supervision, and fine-tuning.
• To highlight the role of explainable AI (XAI) in RNA discovery.

Main Methods:

• Review of existing literature and methodologies in RNA foundation models.
• Explanation of self-supervised learning strategies for RNA sequence data.
• Discussion of fine-tuning approaches for task-specific applications.
• Integration of explainable AI (XAI) for model interpretability.

Main Results:

• Foundation models capture intricate RNA sequence-structure-function relationships.
• Explainable AI (XAI) transforms black-box models into discovery tools.
• Identification of candidate cis-regulatory elements and structural motifs is enabled.

Conclusions:

• RNA foundation models offer a powerful framework for decoding RNA biology.
• Advancements in multimodal data integration promise further discoveries.
• XAI is crucial for translating AI predictions into biological insights.