On the performance of adaptive preprocessing technique in analyzing high-dimensional censored data.

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• High-dimensional censored data, common in genomics (e.g., microarray data), requires robust preprocessing for stability.
• Traditional two-stage variable selection and inference methods for such data are susceptible to circularity bias due to residual noise.
• Addressing noise and collinearity is crucial for reliable inferential analysis in high-dimensional settings.

Purpose of the Study:

• To propose an adaptive preprocessing technique for high-dimensional censored data.
• To mitigate circularity bias inherent in traditional two-stage inferential analyses.
• To enhance variable selection stability and accuracy in the presence of noise and collinearity.

Main Methods:

• The proposed technique integrates sure independence screening (SIS) for initial variable selection.
• It combines SIS with established high-dimensional methods like elastic net variants (elastic net, adaptive elastic net, weighted elastic net, elastic net-AFT) and greedy methods (TCS, PC-simple).
• All methods are implemented within the framework of accelerated lifetime models.

Main Results:

• The adaptive preprocessing technique demonstrates effectiveness in reducing circularity bias.
• It successfully addresses collinearity issues between relevant and irrelevant covariates.
• Performance is validated through simulation studies and analysis of real microarray data (mantle cell lymphoma).

Conclusions:

• The proposed adaptive preprocessing method offers a more stable and less biased approach for variable selection and inference with high-dimensional censored data.
• This technique is particularly beneficial for complex biological datasets like microarrays.
• It provides a valuable tool for researchers dealing with noisy, high-dimensional, and censored outcomes.