PSRR

PSRR predicts regulatory interactions between small molecules and microRNAs (miRNAs), classifying interactions as up-regulation or down-regulation to identify potential small-molecule modulators of miRNA activity.

Key Features:

• Data source: Positive small molecule–miRNA regulation pairs were sourced from the SM2miR database.
• Negative sample generation: Negative or unknown regulation pairs were generated by analyzing structural similarities among small molecules.
• Datasets: Two datasets were produced: Dataset1 for up-regulation pairs and Dataset2 for down-regulation pairs.
• Feature types: Predictions use structural and sequence-based features extracted from small molecules and miRNAs.
• Machine learning models: Models were trained on SM2miR-derived datasets using machine learning, with a random forest model used to derive confidence scores.
• Prediction output: The methodology yields classified regulatory outcomes (up- or down-regulation) accompanied by confidence scores from the random forest model.
• Validation: Predictive performance was validated experimentally and by comparing predicted binding affinities to results from molecular docking simulations.

Scientific Applications:

• Prediction of miRNA regulation: Identification of small molecules that potentially up-regulate or down-regulate target miRNAs based on learned structural and sequence relationships.
• Oncogenic miRNA case study: Application to oncogenic miRNAs from endometrial carcinoma produced top predictions that were experimentally validated and showed binding-affinity agreement with molecular docking.

Methodology:

Positive regulation pairs were obtained from SM2miR; negative/unknown pairs were generated by structural similarity analysis among small molecules, producing Dataset1 (up-regulation) and Dataset2 (down-regulation); structural and sequence-based features were used to train machine learning models on these SM2miR-derived datasets, with a random forest providing confidence scores; predicted binding affinities were compared to molecular docking simulations.