APCanalysis: an R package for identifying active factors using the APC method

Area of Science:

• Statistics
• Experimental Design
• Bioinformatics

Background:

• Unreplicated two-level designs, such as Plackett-Burman designs (PBDs) and fractional factorial designs (2^(k-p)), are crucial for screening experiments.
• Classical statistical methods like t-tests are inadequate for saturated models in these designs due to insufficient degrees of freedom.
• Controlling error rates, including the individual error rate (IER) and experiment-wise error rate (EER), is essential for reliable identification of significant effects.

Purpose of the Study:

• To introduce APCanalysis, a novel R package implementing the All Possible Comparisons (APC) method for statistical screening experiments.
• To provide a user-friendly tool for objective identification of significant effects in complex experimental designs, controlling false positive rates.
• To extend error control beyond IER and EER to include the false discovery rate (FDR) using an advanced penalty algorithm.

Main Methods:

• Implementation of the APC method within the APCanalysis R package, utilizing a tailored AIC-type model selection criterion (APC-criterion).
• Application of an advanced penalty algorithm to extend error control to the false discovery rate (FDR).
• Support for various experimental designs including PBDs, full factorial (2^k), resolution-V (2^(k-p)V), resolution-IV (2^(k-p)IV), and resolution-III (2^(k-p)III) designs.

Main Results:

• The APC-criterion reliably identifies active factors across different experimental designs while adhering to user-specified error thresholds (IER, EER, FDR).
• Simulations and real-world data examples demonstrate the effectiveness and accuracy of the APCanalysis package.
• Benchmarking against Lenth's method shows strong agreement in screening power and accuracy, validating the APC-criterion's performance.

Conclusions:

• APCanalysis offers a robust and user-friendly solution for identifying significant effects in screening experiments using unreplicated two-level designs.
• The package effectively controls various error rates, providing reliable results for complex experimental data analysis.
• APCanalysis is a valuable tool for researchers seeking accurate and statistically sound methods for experimental screening, available on CRAN.