Riboexp

Riboexp models ribosome density on mRNA transcripts using deep reinforcement learning to study translation elongation.

Key Features:

• Deep Reinforcement Learning-Based Framework: Employs deep reinforcement learning to model the complex dynamics of ribosome distribution on mRNA transcripts.
• Policy Network for Feature Selection: Uses a policy network to perform context-dependent feature selection from sequence data.
• Superior Predictive Performance: Outperforms existing methods by up to 5.9% in per-gene Pearson correlation across datasets from three species.
• Informative Sequence Feature Identification: Identifies more informative sequence features for ribosome density prediction than traditional attribution methods.

Scientific Applications:

• Translation Dynamics Studies: Predicts ribosome densities to analyze factors influencing translation elongation.
• Protein Synthesis Optimization: Applied to codon optimization, yielding approximately a 31% increase in protein production compared with previous methods.

Methodology:

A neural network is trained using deep reinforcement learning that incorporates a policy network for context-dependent feature selection to predict ribosome densities across transcripts and identify key sequence features.