In stage 1 of the workflow, we performed an extensive literature search to compile 17134 published research articles which are likely to contain information on EDCs (Figure 1). Firstly, we mined PubMed to filter abstracts on EDCs which led to 19723 research articles. Secondly, we compiled research articles from three existing resources on EDCs, namely, the World Health Organization (WHO) report, The Endocrine Disruption Exchange (TEDX) and EDCs Databank which resulted in 337, 1166 and 456 published literature that are likely to contain information on EDCs, respectively. Subsequently, we manually filtered the compiled abstracts from the above mentioned resources for the presence of keywords related to EDCs, and this filtration led to 17134 research articles at the end of stage 1.

In stage 2 of the workflow, we screened the 17134 research articles from stage 1 to select studies based on in vivo or in vitro experiments in humans or rodents (Figure 1). Here, we have excluded human epidemiological studies and published studies where receptor-based binding assays or in silico methods were employed to infer potential endocrine disruption. The filtration based on study type and test organism led to a subset of 3996 research articles at the end of stage 2.

In stage 3 of the workflow, we gathered the set of chemicals tested for potential endocrine disruption in any of the 3996 research articles from stage 2. Moreover, we also gathered information on the 2D structure of each tested chemical (Figure 1). Note that we have omitted any tested chemical in the 3996 research articles which could not be mapped to its 2D structure using standard chemical databases. At the end of stage 3, we compiled 2047 chemicals along with their 2D structures that were tested for endocrine disruption in humans or rodents in at least one of the filtered research articles from stage 2.

In stage 4 of the workflow, we identify EDCs among the 2047 chemicals compiled in stage 3 by assessing the significance of observed effects in published experiments for endocrine disruption upon exposure in humans or rodents. Prior to this assessment of supporting evidence for endocrine disruption upon chemical exposure, we excluded a tested chemical or its published experiment based on the following criteria (Figure 1):

• Chemical is a natural hormone.
• Chemical was tested as part of a mixture in the published experiment. This criterion reflects our choice to include chemicals which as single entities can cause endocrine disruption upon exposure.
• Chemical was tested for therapeutic relevance in the published experiment.

Moreover, we excluded published experiments which contain evidence for endocrine disruption upon chemical exposure in an in vitro rodent system. Since the observed effects in an in vitro rodent system do not adequately reflect the complexities observed in humans, the last criterion omits such evidence in the published literature (Figure 1).

Next, we evaluated each chemical that passed the above-mentioned criteria, for the level of supporting evidence for endocrine disruption in humans or rodents upon exposure, based on published experiments contained in the filtered research articles. For this evaluation, we manually compiled the observed effects upon exposure of each chemical in associated published experiments in humans or rodents. A published experiment in humans or rodents is considered as strong supporting evidence for endocrine disruption by a chemical if the chemical upon exposure leads to observed effects or endpoints related to endocrine-specific perturbations such as changes in morphology, physiology, growth, reproduction, development and lifespan. Thereafter, if a chemical has at least one published experiment with strong supporting evidence for endocrine disruption upon exposure, then it is identified as a endocrine disrupting chemical (EDC) in stage 4 of the workflow. At the end of stage 4, we identified 792 EDCs with strong supporting evidence of systems-level endocrine-mediated perturbations in published literature spanning 2218 research articles.

In order to standardize the reported evidence for an EDC, we undertook an extensive manual effort to unify the biological terms used to describe the observed effects or endpoints related to endocrine-specific perturbations in published experiments upon chemical exposure. This has led to the compilation of 609 endocrine-mediated endpoints from published experiments in supporting literature for 792 EDCs. Motivated by the major biological processes controlled by the human endocrine system, we have classified the 609 endocrine-mediated endpoints into 7 systems-level perturbations, namely,

• Reproductive endocrine-mediated perturbations (RT)
• Developmental endocrine-mediated perturbations (DT)
• Metabolic endocrine-mediated perturbations (MT)
• Immunological endocrine-mediated perturbations (IT)
• Neurological endocrine-mediated perturbations (NT)
• Hepatic endocrine-mediated perturbations (HT)
• Endocrine-mediated cancer (CT)

We highlight that future studies and toxicological databases can leverage our comprehensive list of 609 endocrine-mediated endpoints and their categorization into 7 systems-level perturbations while reporting or documenting the adverse effects related to endocrine disruption from experiments related to chemical exposure. Hence, this work also contributes towards development of a unified biological vocabulary to describe toxicity profiles of chemicals.