MINT: Mutual Information Based Transductive Feature Selection for Genetic Trait Prediction.

Area of Science:

• Genomics
• Statistical Genetics
• Bioinformatics

Background:

• Whole genome prediction of complex traits is crucial for breeding and genetic epidemiology.
• High-density genotyping data presents a high-dimensional challenge (curse of dimensionality), leading to computational inefficiency and poor model performance due to overfitting and uninformative features.

Purpose of the Study:

• To propose a novel transductive feature selection method, MINT, to address the curse of dimensionality in genomic prediction.
• To evaluate MINT's performance against state-of-the-art methods on genetic trait prediction tasks.

Main Methods:

• Developed MINT, a transductive feature selection method.
• MINT is based on the Max-Relevance and Min-Redundancy (MRMR) criterion.
• Applied MINT to genetic trait prediction problems.

Main Results:

• MINT effectively addresses the curse of dimensionality in genomic prediction.
• MINT demonstrates superior performance compared to the inductive MRMR method in genetic trait prediction.
• The proposed method offers improved computational efficiency and predictive accuracy.

Conclusions:

• MINT is a more effective feature selection approach than existing inductive methods for genomic prediction.
• This method has significant implications for advancing plant and animal breeding and genetic epidemiology.
• MINT provides a robust solution for handling high-dimensional genomic data.