Implementing ReliefF filters to extract meaningful features from genetic lifetime datasets

Area of Science:

• Bioinformatics
• Computational Biology
• Genetics

Background:

• Survival data analysis is crucial for understanding genetic exposure's relation to event occurrence.
• Complex diseases necessitate bioinformatics tools for modeling non-linear interactions in lifetime datasets.
• Existing tools face computational challenges with large-scale datasets.

Purpose of the Study:

• To develop and evaluate bioinformatics tools for feature selection in lifetime datasets.
• To address computational costs associated with survival dimensionality reduction algorithms.
• To estimate attribute quality for extracting relevant features and reducing dataset size.

Main Methods:

• Modified ReliefF algorithm (sReliefF) to handle censored survival data.
• Incorporated reclassification and weighting schemes to compensate for information loss.
• Generated synthetic lifetime datasets with varying heritability and censorship levels for evaluation.

Main Results:

• sReliefF methods demonstrated efficient dataset size reduction.
• Univariate selection methods performed comparably to random chance.
• sReliefF effectively identified causative attribute pairs in simulated data.

Conclusions:

• sReliefF approaches significantly reduce computational costs for large-scale survival data.
• These methods enhance the classification performance of algorithms modeling high-order interactions.
• The developed techniques are valuable for analyzing complex human diseases with right-censored data.