TREAT

TREAT facilitates discovery, screening, and optimization of therapeutic RNAs, with emphasis on integrating regulatory relationships and AI-driven design for coding and noncoding RNA therapeutics.

Key Features:

• Extensive regulatory network integration: Includes 43,087,953 regulatory relationships between coding and noncoding genes across 81 biological networks under various physiological conditions.
• Advanced target screening algorithms: Applies graph representation learning and graph-based methods including Random Walk Diffusions, Topological Degree, and PageRank for disease-relevant target screening.
• AI-driven design and optimization: Performs multi-objective optimization by stratifying features into local and global categories to narrow the search space for designing large RNAs and interfering RNAs.
• Local feature assessment: Evaluates local features within fixed-length or dynamic-length bins to capture position-specific sequence properties.
• Global feature assessment: Uses global assessments inspired by AI language models for protein sequences to evaluate sequence-wide properties.
• Noncoding RNA focus: Explicitly considers noncoding RNAs and their regulatory roles in disease mechanisms as therapeutic targets.

Scientific Applications:

• Target identification and prioritization: Screen and prioritize disease-relevant coding and noncoding RNA targets using large-scale regulatory networks and graph-based algorithms.
• RNA design and optimization: Design and optimize large RNAs and interfering RNAs through AI-driven multi-objective optimization that balances local and global sequence features.
• Mechanistic investigation of noncoding RNAs: Analyze regulatory relationships of noncoding RNAs across physiological conditions to inform therapeutic hypotheses and intervention strategies.

Methodology:

Targeted RNA screening using graph-based algorithms (graph representation learning, Random Walk Diffusions, Topological Degree, PageRank), followed by AI-driven multi-objective optimization that stratifies features into local (fixed-length or dynamic-length bins) and global (language-model-inspired) assessments to design and refine RNA candidates and address immunogenicity, instability, and translational inefficiency.