ilp-datasets / tox21 /README.md
andrewcropper's picture
Add files using upload-large-folder tool
cb1dd83 verified
|
Raw
History Blame Contribute Delete
3.35 kB

Tox21

Molecular toxicity dataset across 12 nuclear receptor and stress response pathway assays. Each molecule is represented by its atom and bond structure; the learning task is to predict whether a molecule is active (toxic) in a given assay.

Tasks: 12 tasks (in <task>/train/)

Task Description
nr_ahr Aryl hydrocarbon receptor activation
nr_ar Androgen receptor activation
nr_ar_lbd Androgen receptor ligand-binding domain activation
nr_aromatase Aromatase enzyme inhibition
nr_er Estrogen receptor activation
nr_er_lbd Estrogen receptor ligand-binding domain activation
nr_ppar_gamma PPAR-gamma (peroxisome proliferator-activated receptor gamma) activation
sr_are Antioxidant response element pathway activation
sr_atad5 ATAD5 genotoxicity stress response
sr_hse Heat shock response element pathway activation
sr_mmp Mitochondrial membrane potential disruption
sr_p53 p53 tumour suppressor pathway activation

Head relation: f(Mol) — true if molecule Mol is active (positive) in the assay.

Background relations:

Relation Arity Description
atom(Mol, Atom, Element) 3 Atom Atom in molecule Mol has element symbol Element
bond(Mol, Atom1, Atom2, BondType) 4 Atoms Atom1 and Atom2 in molecule Mol are connected by a bond of type BondType
bond_type_single(BondType) 1 BondType is a single bond
bond_type_double(BondType) 1 BondType is a double bond
bond_type_triple(BondType) 1 BondType is a triple bond
bond_type_aromatic(BondType) 1 BondType is an aromatic bond
element_c(Element) 1 Element is carbon (C)
element_n(Element) 1 Element is nitrogen (N)
element_o(Element) 1 Element is oxygen (O)
element_s(Element) 1 Element is sulfur (S)
element_* 1 One relation per element; 51 element types total (Ag, Al, As, Au, B, Ba, Be, Bi, Br, C, Ca, Cd, Cl, Co, Cr, Cu, Dy, F, Fe, Gd, Ge, H, Hg, I, In, K, Li, Mg, Mn, Mo, N, Na, Nd, Ni, O, P, Pb, Pd, Pt, S, Sb, Se, Si, Sn, Sr, Ti, Tl, V, Yb, Zn, Zr)

Dataset statistics: 7,831 molecules total; each task directory contains only molecules that were tested in that assay (many molecules have no label for a given assay). Positive (active) counts range from 186 (nr_ppar_gamma) to 942 (sr_are).

Task statistics

task_name num_bk_facts num_bk_relations num_pos_examples num_neg_examples
nr_ahr 240887 57 768 5781
nr_ar 272913 57 309 6956
nr_ar_lbd 245942 57 237 6521
nr_aromatase 208008 57 300 5521
nr_er 221474 57 793 5400
nr_er_lbd 256146 57 350 6605
nr_ppar_gamma 228033 57 186 6264
sr_are 194501 57 942 4890
sr_atad5 259036 57 264 6808
sr_hse 222174 57 372 6095
sr_mmp 207909 57 918 4892
sr_p53 247030 57 423 6351

Source: Tox21 Data Challenge (2014). OGB dataset ogbg-moltox21. Hu, Z. et al. "Open Graph Benchmark: Datasets for Machine Learning on Graphs." NeurIPS, 2020. https://ogb.stanford.edu/