Classification of riboswitch sequences using k-mer frequencies

Area of Science:

• Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• Riboswitches are non-coding RNAs that regulate gene expression through structural changes.
• Their potential in biomedical applications like drug targets and biosensors is significant.
• Accurately identifying diverse and structurally complex riboswitches remains a challenge.

Purpose of the Study:

• To investigate the classification of 16 riboswitch families using supervised learning.
• To develop a sequence-based feature set for riboswitch identification.
• To compare the performance of various machine learning classifiers.

Main Methods:

• Generated a reduced, sequence-based feature set for riboswitches.
• Employed supervised learning algorithms: Support Vector Machine, Random Forest, Naive Bayes, J48, and HyperPipes.
• Tested classifier performance on independent datasets.

Main Results:

• The best multi-class classifier achieved high F-measure scores (0.996 training, 0.966 testing).
• The proposed sequence-based feature set enabled accurate riboswitch discrimination.
• Classifiers demonstrated competitive performance compared to BLAST.

Conclusions:

• Supervised learning with sequence-based features can accurately classify riboswitch families.
• This approach offers an effective method for identifying metabolite-binding RNA switches.
• The findings advance the potential for riboswitches in biomedical applications.