One-step extrapolation of the prediction performance of a gene signature derived from a small study

Objective: Microarray-related studies often involve a very large number of genes and small sample size. Cross-validating or bootstrapping is therefore imperative to obtain a fair assessment of the prediction/classification performance of a gene signature. A deficiency of these methods is the reduced training sample size because of the partition process in cross-validation and sampling with replacement in bootstrapping. To address this problem, we aim to obtain a prediction performance estimate that strikes a good balance between bias and variance and has a small root mean squared error. Methods: We propose to make a one-step extrapolation from the fitted learning curve to estimate the prediction/classification performance of the model trained by all the samples. Results: Simulation studies show that the method strikes a good balance between bias and variance and has a small root mean squared error. Three microarray data sets are used for demonstration. Conclusions: Our method is advocated to estimate the prediction performance of a gene signature derived from a small study.