Conditional deep learning model reveals translation elongation determinants during amino acid deprivation

Area of Science:

• Molecular Biology
• Computational Biology
• Genomics

Background:

• Translation elongation is crucial for cellular homeostasis.
• Dysregulation of translation is linked to diseases and metabolic disorders.
• Understanding intragenic translation heterogeneity under amino acid deprivation is vital for therapeutic development.

Purpose of the Study:

• To develop an accurate and explainable computational framework for predicting ribosome footprint profiles.
• To identify sequence-based determinants of ribosome stalling during amino acid deprivation.
• To investigate the impact of intragenic variations on translation regulation.

Main Methods:

• Development of Riboclette, a conditional deep learning model with a dual output head.
• Utilizing mRNA sequence as input to predict genome-wide ribosome footprint profiles.
• Application of interpretability methods and in silico perturbation experiments.

Main Results:

• Riboclette accurately predicts ribosome footprint profiles across six amino acid deprivation conditions.
• Identified codons for deprived, poly-basic, and negatively charged amino acids as key drivers of ribosome stalling.
• Extracted motif-level drivers of ribosome stalling, explaining positional effects.

Conclusions:

• Riboclette provides an accurate and explainable method for studying translation elongation regulation.
• The findings enhance understanding of disease mechanisms related to translation disruption.
• This framework can aid in the development of novel therapeutics targeting translation.