This is a tool to screen likely toxicological effects of novel molecules trained on aggregated toxicity data from more than 10,000 chemicals
This model accepts a SMILES string as an input. You can use this free PubMed tool to translate your structure into a SMILES string
Undestanding Output
A "1" for a given endpoint indicates that a toxicologically relevant interaction is likely
A "0" indicates that the model expects the molecule to be benign relative to the given endpoint
Endpoints Included
Nuclear Receptor Assays test the interaction of compounds with various proteins within cells that are responsible for sensing steroid and thyroid hormones and certain other molecules. Interactions with these receptors can indicate potential endocrine-disrupting effects of the compounds.
'NR.AhR': Nuclear Receptor Aryl hydrocarbon Receptor assay.
'NR.AR': Nuclear Receptor Androgen Receptor assay.
'NR.AR.LBD': Androgen Receptor Ligand Binding Domain assay.
'NR.Aromatase': Aromatase enzyme assay.
'NR.ER': Nuclear Receptor Estrogen Receptor assay.
'NR.ER.LBD': Estrogen Receptor Ligand Binding Domain assay.
'NR.PPAR.gamma': Nuclear Receptor Peroxisome Proliferator-Activated Receptor Gamma assay.
Stress Response Assays measure the response of cells to oxidative stress, heat shock, and DNA damage. They help in understanding how a compound might induce cellular stress or damage at a molecular level.
'SR.ARE': Antioxidant Response Element assay.
'SR.ATAD5': ATAD5 assay, possibly relating to a DNA damage response.
'SR.HSE': Heat Shock Element response assay.
'SR.MMP': Mitochondrial Membrane Potential assay.
'SR.p53': p53 assay, related to the tumor suppressor protein p53.
Model Construction and Details
Input strings are decomposed into 209 unique features, grouped broadly into these categories:
- Topological Descriptors are based on the molecular graph, these values describe the topology of the molecule. They do not consider their positions in three dimensional space . They can include path counts, cluster counts, and other graph-based metrics.
- Geometric Descriptors capture the spatial geometry of molecules, including aspects like molecular volume, shape, and surface area.
- Electronic Descriptors describe electronic properties of molecules, such as their polarizability, electron distribution, and potential for interaction with other molecules.
- Constitutional Descriptors are the simplest form of descriptors and include counts of certain atom types or functional groups within a molecule
- Molecular Property Descriptors are directly calculated from the molecular structure and include properties such as molecular weight
- Pharmacophore Features describe the presence of specific pharmacophoric elements that are important for drug activity, such as hydrogen bond donors and acceptors, nitro groups, and oxygens.
- E-state Indices describe the electronic state and topological environment of atoms within a molecule
- Molecular Fragment Counts count specific molecular fragments or substructures within a molecule,
Model Analysis
Model is optimized to minimize false positives. Throwing out a valuable candidate is a very bad outcome in this space. If you'd like to experiment with models optimized for different outcomes please contact me directly.
Deeply curious about the methodology, logic, back end, detailed evaluation metrics? Head to the project's github and read the model analysis for full details.