ConsAlign: simultaneous RNA structural aligner based on rich transfer learning and thermodynamic ensemble model of...

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Biology

Background:

• RNA alignment and folding (AF) is crucial for understanding structural homology in RNA homologs.
• Developing effective scoring parameters for simultaneous AF (SAF) is challenging due to computational expense.

Purpose of the Study:

• To develop a machine learning method for improved RNA simultaneous alignment and folding (SAF) scoring.
• To implement a tool that enhances AF quality using learned scoring parameters.

Main Methods:

• Developed ConsTrain, a gradient-based machine learning method for SAF scoring.
• Implemented ConsAlign, a SAF tool utilizing ConsTrain's parameters.
• Employed transfer learning and ensemble modeling (ConsTrain + thermodynamic model) for enhanced AF quality.

Main Results:

• ConsAlign demonstrated competitive AF prediction quality compared to existing tools.
• The method maintained comparable running times.
• Achieved rich SAF scoring through machine learning.

Conclusions:

• ConsTrain and ConsAlign offer an effective machine learning-based solution for RNA SAF.
• The developed tools provide a computationally efficient approach to improve RNA homology detection.
• Code and data are publicly available for further research.