Adjustment of spurious correlations in co-expression measurements from RNA-Sequencing data

Abstract

Gene co-expression measurements are widely used in computational biology to identify coordinated expression patterns across a group of samples, which may indicate that these genes are controlled by the same transcriptional regulatory program, or involved in common biological processes. Gene co-expression is generally estimated from RNA-Seq data, which are generally normalized to remove technical variability. Here, we find and demonstrate that certain normalization methods, in particular quantile-based methods, can introduce false-positive associations between genes, and that this can consequently hamper downstream co-expression network analysis. Quantile-based normalization can, however, be extremely powerful. In particular when preprocessing large-scale heterogeneous data, quantile-based normalization can be applied to remove technical variability while maintaining global differences in expression for samples with different biological attributes. We therefore developed CAIMAN, a method to correct for false-positive associations that may arise from normalization of RNA-Seq data. CAIMAN utilizes a Gaussian mixture model to fit the distribution of gene expression and to adaptively select the threshold to define lowly expressed genes, which are prone to form false-positive associations. Thereafter, CAIMAN corrects the normalized expression for these genes by removing the variability across samples that might lead to false-positive associations. Moreover, CAIMAN avoids arbitrary gene filtering and retains associations to genes that only express in small subgroups of samples, highlighting its potential future impact on network modeling and other association-based approaches in large-scale heterogeneous data.

The Network Zoo: a multilingual package for the inference and analysis of biological networks

“…We accomplished this by standardizing the implementations of software tools built on a common conceptual framework, in line with recent similar efforts [68, 69], and building an ecosystem for sharing of use cases, hosting networks, and continued development and maintenance which is essential for software accuracy [70]. We will continue to expand netZoo (Figure S4), adding new methods [71–73] and improving implementations of the existing tools, as well as building interfaces to allow methods to be combined appropriately. We welcome community participation in methods development and are committed to the broad use of the tools available within netZoo.…”

The Network Zoo: a multilingual package for the inference and analysis of gene regulatory networks

“…We will continue to expand netZoo (Additional file 1: Fig. S5) particularly for single-cell genomics, adding new methods [88][89][90] and improving implementations of existing methods, as well as building interfaces to allow methods to be combined appropriately. We will also continue to leverage the codebase to add new components in the ecosystem of online tools we developed to further aid users and developers in hosting genome-scale networks and running complex analyses on the cloud for their own investigations.…”

Reconstructing Sample-Specific Networks using LIONESS

“…Because LIONESS can be applied to very different types of network reconstruction algorithms and data, we recommend that the user explores how the data and data preprocessing may affect the output networks, as well as identify the downstream network analysis approach that is most appropriate to answer their question of interest. These decisions are critical to gaining important biological insights, as we and others have in our previous applications of the method (Kuijjer et al, 2019a; Sonawane et al, 2019; Fagny et al, 2020; Lopes-Ramos et al, 2018, 2020, 2021, Pham et al, 2021).…”

“…In principle, any method that corrects for technical artifacts in the data and that makes expression data comparable across samples can be used. However, keep in mind that the technique of normalization selected could have an effect on the output co-expression values (Hsieh et al, 2021). It is therefore important to evaluate the effect the chosen preprocessing technique may have on the chosen network reconstruction algorithm to be used with LIONESS.…”

“…On occasion, the LIONESS equation may result in edge weights that are not normally distributed across samples, in which case it may be more appropriate to use nonparametric statistics when comparing single-sample edge weights. However, in our experience, node degree is often normally distributed across samples and can generally be compared with, for example, Student’s t-statistic or linear regression-based approaches (Lopes-Ramos et al, 2018, 2020, 2021).…”