A framework for list representation, enabling list stabilization through incorporation of gene exchangeabilities

Analysis of multivariate data sets from e.g. microarray studies frequently results in lists of genes which are associated with some response of interest. The biological interpretation is often complicated by the statistical instability of the obtained gene lists with respect to sampling variations, which may partly be due to the functional redundancy among genes, implying that multiple genes can play exchangeable roles in the cell. In this paper we use the concept of exchangeability of random variables to model this functional redundancy and thereby account for the instability attributable to sampling variations. We present a flexible framework to incorporate the exchangeability into the representation of lists. The proposed framework supports straightforward robust comparison between any two lists. It can also be used to generate new, more stable gene rankings incorporating more information from the experimental data. Using a microarray data set from lung cancer patients we show that the proposed method provides more robust gene rankings than existing methods with respect to sampling variations, without compromising the biological significance.

💡 Research Summary

The paper addresses a pervasive problem in high‑dimensional genomic studies: gene lists derived from microarray or similar experiments are often unstable with respect to sampling variation. Traditional approaches mitigate this instability by resampling, cross‑validation, or adding stability‑based weights, yet they ignore the fact that many genes are functionally redundant—different genes can fulfill interchangeable roles in cellular pathways. To capture this redundancy, the authors import the probabilistic notion of exchangeability from the theory of random variables. Two genes are said to be exchangeable if swapping them does not materially change the statistical summary of the data. By estimating pairwise exchangeability scores, the authors construct an “exchangeability matrix” that quantifies how interchangeable each gene pair is.

The methodology proceeds in several steps. First, a bootstrap procedure is applied to the original expression matrix to generate many resampled datasets. For each bootstrap replicate a gene ranking (e.g., based on differential expression) is computed. The frequency with which two genes appear together in similar rank positions across replicates is used as an empirical estimate of their exchangeability. This yields a symmetric matrix E where Eij ∈