Large-Scale Labeling and Assessment of Sex Bias in Publicly Available Expression Data

Abstract

Women are at more than 1.5-fold higher risk for clinically relevant adverse drug events. While this higher prevalence is partially due to gender-related effects, biological sex differences likely also impact drug response. Publicly available gene expression databases provide a unique opportunity for examining drug response at a cellular level. However, missingness and heterogeneity of metadata prevent large-scale identification of drug exposure studies and limit assessments of sex bias. To address this, we trained organism-specific models to infer sample sex from gene expression data, and used entity normalization to map metadata cell line and drug mentions to existing ontologies. Using this method, we infer sex labels for 450,371 human and 245,107 mouse microarray and RNA-seq samples from refine.bio. Overall, we find slight female bias (52.1%) in human samples and (62.5%) male bias in mouse samples; this corresponds to a majority of single sex studies, split between female-only and male-only (33.3% vs 18.4% in human and 31.0% vs 30.4% in mouse respectively). In drug studies, we find limited evidence for sex-sampling bias overall; however, specific categories of drugs, including human cancer and mouse nervous system drugs, are enriched in female-only and male-only studies respectively. Our expression-based sex labels allow us to further examine the complexity of cell line sex and assess the frequency of metadata sex label misannotations (2-5%). We make our inferred and normalized labels, along with flags for misannotated samples, publicly available to catalyze the routine use of sex as a study variable in future analyses.

Comprehensive Assessment of Smoking and Sex Related Effects in Publicly Available Gene Expression Data

“…For smoking history studies, 34.5% of samples and 38.8% of studies were missing metadata sex labels; this is much lower than seen across all human studies and samples (e.g. 70.7% of human microarray samples are missing sex labels (Flynn, Chang, and Altman 2021)). The higher fraction of sex labels in smoking datasets may be related to the fact that smoking status is included, so sex is additionally likely to be recorded as a covariate.…”

“…We additionally sought to examine sex bias overall in smoking-related studies. We focused on the 139 (out of 176) smoking history studies that were included in the refine-bio database by inferring sex labels from gene expression data using our previously published method (Flynn, Chang, and Altman 2021). For smoking history studies, 34.5% of samples and 38.8% of studies were missing metadata sex labels; this is much lower than seen across all human studies and samples (e.g.…”

“…Our previously developed method for logistic regression-based models for sex labeling (Flynn, Chang, and Altman 2021) were trained on normalized data from the refine-bio database (Greene et al n.d). This database consists of over 14,000 human studies from GEO, ArrayExpress, and SRA; however, it is not complete.…”

“…For application at scale, we restricted our assessment of sex bias to these 139 studies. As in (Flynn, Chang, and Altman 2021), we grouped studies into the following categories based on the sample sex labels: 1. Unlabeled: studies with either less than half of their samples labeled (for studies with up to 60 samples) or less than 30 samples labeled (for studies with more than 60 samples) 2.…”

“…liver (Zhang et al 2011), blood (Bongen et al 2019), brain (Trabzuni et al 2013)) have found hundreds of sex differentially expressed (DE) genes. Additionally, multiple methods (Buckberry et al 2014; Ellis et al 2018; Giles et al 2017; Toker, Feng, and Pavlidis 2016; Flynn, Chang, and Altman 2021) have been developed for inferring sex labels from gene expression data, leveraging the highly sexually dimorphic expression of X and Y chromosome genes. Smoking status also has a substantial impact on gene expression: previous studies have identified hundreds of DE genes between smokers and non-smokers in blood (Charlesworth et al 2010; Na et al 2015; Huan et al 2016), airway epithelium (Chen Xi Yang et al 2019; Boelens et al 2009), lung (Landi et al 2008; He et al 2018), and other tissues (Port et al 2004; Na et al 2015; Tsai et al 2018).…”