Analysis of microarray right-censored data through fused sliced inverse regression

Area of Science:

• Statistics
• Machine Learning
• Bioinformatics

Background:

• Sufficient dimension reduction (SDR) aims to simplify regression models by projecting high-dimensional predictors onto a lower-dimensional space.
• Sliced inverse regression (SIR) is a popular SDR method but is sensitive to the number of slices used.
• High-dimensional data, such as microarray data, presents unique challenges for traditional regression techniques.

Purpose of the Study:

• To introduce and evaluate a fused sliced inverse regression (SIR) method for sufficient dimension reduction.
• To address the sensitivity of SIR to the number of slices by integrating results from multiple slice counts.
• To demonstrate the practical advantages of fused SIR in analyzing large-p-small-n regression problems with right-censored, high-dimensional data.

Main Methods:

• A fused approach is proposed that combines kernel matrices from SIR applied with varying numbers of slices.
• The fused SIR method is applied to a high-dimensional microarray dataset with right-censored regression.
• Model validation techniques are employed to compare the performance of fused SIR against standard SIR.

Main Results:

• The fused SIR method effectively handles the challenges of high-dimensional regression with right-censored data.
• Performance evaluation confirms that fused SIR outperforms standard SIR across all considered slice numbers.
• The proposed method demonstrates practical advantages in real-world applications like microarray data analysis.

Conclusions:

• Fused SIR offers a robust and improved alternative to traditional SIR for sufficient dimension reduction.
• This approach mitigates the critical deficit of slice number sensitivity in standard SIR.
• Fused SIR shows significant promise for analyzing complex, high-dimensional biological and other datasets.