TABLE 2.

In silico tools considered for assessing the ADMET profiles of investigated oligomers.

Computational tool	Predicted activity and accuracy of prediction
admetSAR2.0 is a free available structure-activity relationship database that contains over 210,000 available properties data curated from literature for about 96,000 chemicals. It includes 22 qualitative classification models and 5 quantitative regression models allowing estimations quantitatively described by a probability output (Cheng et al., 2012; Yang et al., 2018). As input data, this tool uses the SMILES formulas of investigated oligomers and we have extracted the probabilities of ADMET properties.	It performs predictions concerning: gastrointestinal absorption (GI) (0.965), plasma protein binding (PPB) (0.668), blood brain barrier permeation (BBB) (0.907), substrate/inhibition of the P-glycoprotein (Pgps/Pgpi) (0.802/0.861), substrates (0.779) or inhibitors (0.855) of the human cytochromes (CYPs) involved in the metabolism of xenobiotics, inhibition of organic-anion-transporting polypeptides OATP1B1 (0.886), OATP1B3 (0.927), OATP2B11 (0.885), multidrug and toxin extrusion protein 1 (MATE1) (0.907) and human organic cation transporter OCT2 (0.808), eye corrosion and/or irritation (0.949/0.963), human Ether-a-go-go-Related Gene (hERG) inhibition (0.804), hepatoxicity (0.833), carcinogenicity (0.896), mutagenicity by Ames test (0.843).
ENDOCRINE DISRUPTOME tool uses the molecular docking approach to predict the interactions between the investigated chemicals and 12 human nuclear receptors (Kolsek et al., 2014). As input data we have used the SMILES formulas of investigated oligomers and extracted data considering the color coded table that was displayed when the molecular docking was finished: green for compounds reflecting a low probability of interacting with the nuclear receptors, orange for compounds revealing a mean probability of interacting with the nuclear receptors and red for compounds revealing a high probability of affecting the nuclear receptors.	It predicts interactions with (0.780): androgen receptor (AR)—agonistic and antagonistic interactions, estrogen receptors (ER) α and β, glucocorticoid receptor (GR)—agonistic and antagonistic interactions, liver X receptors (LXR) α and β, peroxisome proliferator activated receptors (PPAR) α, β/δ, and γ, retinoid X receptor (RR) α, thyroid receptors (TR) α and β.
Pred-Skin 3.0 is a web-server allowing predictions concerning skin sensitization potential of chemicals based on QSAR models (Braga et al., 2017; Alves et al., 2018a). SMILES formulas have been used as input data and the probability of oligomers to illustrate skin sensitization or non-sensitization potentials, respectively, the probability maps illustrating the fragments contributions toward skin sensitization potential were retrieved.	It performs predictions based on five sensitization assay: in vivo (murine local lymph node assay, LLNA, accuracy 0.70–0.84), in chemico (Direct Peptide Reactivity Assay, DPRA, accuracy 0.73–0.76), in vitro (KeratinoSens and human Cell Line Activation Test, H-CLAT, accuracy 0.80–0.86), human repeated insult patch, HRIPT, test and human maximization test, HMT, accuracy 0.70–0.84. There also is a Bayesian consensus model that is generated by averaging the predictions of individual models. Pred-Skin 3.0 also outcomes a probability maps allowing the visualization of the contribution of predicted fragment toward skin sensitization.
Pred-hERG 4.2 is a web tool that builds predictive models of the ability of a chemical compound to inhibit the human Ether-à-go-go Related Gene (hERG) based on the QSAR approach (Braga et al., 2015). SMILES formulas have been used as input data and the probabilities of oligomers to illustrate hERG K+ channel blocking or non-blocking potential, together with the probability maps illustrating the fragments contributions toward hERG blockage potential were retrieved.	The outcome is a binary prediction of hERG non-blocker or blocker (0.80). This tool also delivers a probability maps allowing the visualization of the contribution of predicted fragment toward hERG blockage.
CarcinoPred-EL (Carcinogenicity Prediction using Ensemble Learning methods) is a computational tool used for predictions concerning the carcinogenicity of chemicals ensemble classification models (Zhang et al., 2017). SMILES formulas have been used as input data and the software provides a binary prediction (Yes or No) for possible carcinogenicity of the investigated compounds.	The carcinogenic potential of chemicals is predicted using: Ensemble SVM model (0.691), Ensemble RF model (0.686), Ensemble XGBoost model (0.698).
Toxtree is an open-source application that performs predictions concerning carcinogenicity and mutagenicity by applying a decision tree approach (Patlewicz et al., 2008). As input data we have used the SMILES formulas of investigated oligomers and we have retrieved the predictions (Yes or No) for the carcinogenic and mutagenic potential of investigated compounds.	The carcinogenic and mutagenic potential is predicted (0.70).
PASS (Prediction of Activity Spectra of Substances) is a computational tool that predicts biological activity spectra and toxic/side effects starting to the structural formulae of chemical compounds and using the QSAR approach (Poroikov et al., 2007). This tool also used the SMILES formulas as entry data and provides the probability that the investigated compound to be active for a given adverse biological activity.	PASS has been used to predict toxic and adverse effects (0.95).

Numbers in parantheses represent probabilities of predictions.