Tailored machine learning models for functional RNA detection in genome-wide screens

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• In silico prediction of genetic loci is crucial for understanding nucleotide sequence functions.
• Existing annotation pipelines often lack flexibility and user-defined parameters.

Purpose of the Study:

• To present a software framework for alignment-based machine learning model training and evaluation.
• To offer a flexible alternative to one-size-fits-all in silico annotation pipelines.

Main Methods:

• Developed a software framework for customizable machine learning model generation.
• Applied the framework for a full-genome screen in Drosophila melanogaster.
• Evaluated results against the RNAz program.

Main Results:

• The framework allows for structured generation and evaluation of models based on arbitrary features and input data.
• Demonstrated stable and explainable results in a Drosophila melanogaster genome screen.
• Showcased the framework's utility and flexibility compared to existing tools.

Conclusions:

• The presented framework provides a powerful and adaptable tool for in silico genetic locus prediction.
• It facilitates deeper insights into the biological roles of nucleotide sequences through user-defined, explainable models.