Skip to main content
Contact Info
Bryant Chambers

OpenTox 2022 Virtual Conference 

Searching for LINCS to Stress: literature-linked transcriptomic analysis identifies stress  response active chemical targets, MoAs, and use classes  

Bryant Chambers1*, Danilo Basili2, Laura Word1, Nancy Baker3, Alistair Middleton2, Richard Judson1, Imran  Shah1 

1 Center for Computational Toxicology and Exposure, Office of Research and Development, U.S. Environmental Protection Agency,  Research Triangle Park, North Carolina, USA  

2 Unilever, Safety and Environmental Assurance Centre (SEAC), Colworth Science Park, Sharnbrook, Bedfordshire MK44 1LQ, U.K  3 Leidos, Research Triangle Park, NC, USA  

* Presenting Author  

Adaptive stress response pathways (SRPs) restore homeostasis via transcriptionally regulated processes that  sense deviation from a healthy state. Persistent SRP activation is associated with adverse outcomes (e.g.,  necrosis and apoptosis) and diseases (e.g., type II diabetes). Despite their essential role in maintaining  homeostasis, there is a lack of information about chemicals known to activate SRPs and tools for identifying  them from transcriptomic data. Mining literature to find associations between chemicals and SRPs could help  improve SRP annotation and analysis. Pairwise mutual information (PMI) is an information theory approach that  normalizes cooccurrence by balancing the joint occurrence of two entities against the independent occurrence  of each. Applying PMI to article counts shared between SRPs and chemicals could facilitate annotating SRP  active chemicals to support screening tools. We measured cooccurrence between article counts for SRPs in a  literature database and chemicals in a transcriptomic database to annotate chemicals with SRP activity and  afterward validated those annotations with transcriptomic data. We evaluated six canonical SRPs: DNA damage  (DDR), heat shock (HSR), hypoxia (HPX), metal stress (MSR), oxidative stress (OSR), and unfolded protein  (UPR). First, we calculated the PMI between 4,761 chemicals and the six SRPs to normalize cooccurrence  from article counts. We next hierarchically clustered both measures of cooccurrence for database chemicals  with SRP phrases and then measured the accuracy of chemical-SRP relationships by searching for groups of  chemicals with known functions. We checked for mapping between transcriptomic data and literature-derived  SRP annotations with t-distributed stochastic neighbor embedding (T-SNE) to find clusters of transcriptomic  profiles enriched for SRP annotations. We then quantified SRP bioactivity in transcriptomic data by gene set  enrichment analysis (GSEA) with six built-to-purpose SRP signatures. SRP reference chemicals clustered better  by PMI than article counts; DDR, OSR, UPR, and HSR clusters were visible in PMI dendrograms. PMI clusters  included fewer spurious chemicals, e.g., metabolites, than article count clusters. Recall of hand-validated  classification was better for PMI annotation (88%) compared to article counts (68%). We next built a  transcriptomic test set from strong SRP annotations. T-SNE clustered transcriptomic profiles of this test set  formed regions that mapped to SRP annotations and GSEA of SRP signature scores matched with PMI SRP  annotation (440 chemicals, signature Z score > 4 SD) indicating correspondence between literature and  transcriptomic data. Results suggest that automating balanced SRP annotation is possible. Further, mapping  mechanisms of action (MoAs) supported SRP annotations and agreed with GSEA analysis. We observed  expected MoAs enriched for SRP activity (>10 SD). MoAs such as HSR protein inhibitors (p = e-12) and  intercalating agents (p = 0.04) associated with HSR and DDR assignments, respectively. Importantly, SRPs  were implicated in the activity of injurious agents such as drug-induced liver injury (DILI) chemicals of concern  (p = e-40). These data illustrate a new method to validate transcriptomic signals en masse and identify  reference chemicals while underlining the usefulness of SRPs as measures of diseases.  

Opinions expressed here do not necessarily reflect the opinions of USEPA nor should they be taken as policy or advisement.