circLGB

circLGB employs machine learning to discriminate circular RNAs (circRNAs) from other long non-coding RNAs (lncRNAs) and to predict associated regulatory information.

Key Features:

• Feature integration: Incorporates common sequence-derived features alongside three novel features: Adenosine to Inosine (A-to-I) deamination, A-to-I density, and Internal ribosome entry site.
• Machine learning classifier: Uses a LightGBM classifier and applies feature selection to optimize circRNA versus lncRNA classification.
• Modeled feature types: Represents sequence-based features, graph features, genome context, and regulatory information features for prediction.
• Regulatory information prediction: Predicts regulatory information associated with circRNAs using an ensemble machine learning approach.

Scientific Applications:

• circRNA identification: Improves accuracy of discriminating circRNAs from other lncRNAs to facilitate study of circRNA biogenesis and regulatory roles.
• Interaction analysis: Enables investigation of circRNA interactions with microRNAs, RNA-binding proteins, and transcriptional regulators.

Methodology:

Implements an ensemble machine learning approach that models sequence-based, graph, genome-context, and regulatory-information features, employs a LightGBM classifier with feature selection, and evaluates performance on public and newly constructed datasets.