QSAR Design of Discovery Libraries for Solids Based on QSAR Models 2005 QSAR and rial Science
In silico workflow for assessing skin penetration ... · •for QSAR models, model validation...
Transcript of In silico workflow for assessing skin penetration ... · •for QSAR models, model validation...
Detailed results are shown below for one test set compound. According to Dempster-Shafer theory, each prediction consists of a reported belief and plausibility. The difference between these two is the uncertainty associated with the prediction. The global model suggests fumaric acid may be a moderate sensitizer but the prediction is uncertain. The SN2 carboxylate MoA model predicts non-sensitizer. qWoE prediction is non-sensitizer with probability 44% to 76% [belief plausibility].
A rigorous decision-theory approach based on Dempster-Shafer theory (DST) has been developed to accomplish two key tasks essential for probabilistic modeling:
1) uncertainty analysis
• a quantitative estimate of the uncertainty associated with each prediction is determined based on the reliability of the evidence source
• for QSAR models, model validation process provides reliability metrics
• applicable to binary, multinomial, and ordinal classification models
• can be applied to any source of evidence: QSAR models, structural alerts, in vivo and in vitro assay data, expert opinions, …
2) combination of multiple sources of evidence
• evidence is combined by joining DST basic probability mass structures
• much more rigorous and versatile than simplistic combination-of-evidence strategies commonly used (e.g., consensus or mean-value methods)
• enables quantitative weight-of-evidence (qWOE) strategies for combining multiple and diverse information sources
• combination reduces uncertainty when sources are in general agreement, and increases uncertainty when strong conflicts exist
J. Rathman1,2 C. Yang1,2,3 L. Terfloth3 J. Schwöbel3 A. Mostrag-Szlichtyng1 T. Magdziarz3 A. Tarkhov3
Contact: [email protected] Acknowledgement: www.cosmostox.eu
In silico workflow for assessing skin penetration, irritation,
and sensitization potentials using chemotype-based models and alerts
INTRODUCTION SKIN SUITE COMPUTATIONAL WORKFLOW
ESTIMATING UNCERTAINTY AND
QUANTITATIVE COMBINATION OF EVIDENCE
1Altamira LLC
Columbus OH
TRAINING RESULTS AND PREDICTIONS
Evaluating chemical toxicity instigated by dermal contact requires
addressing exposure (permeation) and chemical reactivity that leads to
irritation or subsequent sensitization via well-known induction/elicitation
immune response mechanisms:
2The Ohio State University
Columbus, OH
3Molecular Networks
Erlangen, Germany
Induction Elicitation
OECD report ENV/JM/MONO(2012)10/PART1
Stratum corneum
Epidermis
Dermis
Lymph nodes
high no
global SNAr SN2 carboxylic
SN2 alkyl halides
Michael acceptors
Schiff base formers
phenols skin
sensitization models
Quantitative weight-of-evidence (qWOE) approach to combine evidence sources and estimate prediction uncertainty
alert?
non-sensitizer
Schiff base formers
Michael acceptors
Aldehyde binding to lysine
a,b-unsaturated carbonyl binding to cystein
SNAr
Aromatic nucleophilic substitution
SN2 e.g. primary-alkyl halides. etc.
X Nu -+ Nu:-
+ X
Smith Pease, C.K., Toxicology, 192, 2003, 1-22
Aptula, et. al. Chem. Res. Toxicol. 19, 2006, 1097
Database
In chemico reactivity database > 400 chemicals
‒ GSH assay
• Michael acceptors (43), SN2 (73)
‒ Methanolate assay
• SNAr (54)
‒ Reaction sites
‒ Reaction rates
Local lymph node assay (LLNA) assay database
‒ Database (> 470 structures): Master table I and II, other publications, OECD toolbox
‒ Study design (OECD TG429): 3H-Methymidine assay, vehicle (>240 used AOO,
ethanol or acetone/H2O)
‒Results: Stimulation index, EC3, LLNA potency
Exposure? • skin permeability database • models (Potts & Guy, Kasting) • QSAR (kp, Jmax, logP, ilogP, …)
query molecule (example: fumaric acid)
Irritant? • corrosion alerts • QSAR (MoA models based on
chemotypes, physchem props)
LLNA TRAINING SET
non weak moderate strong extreme
potency
freq
uen
cy
0
60
12
0 Influential descriptors
(global model):
ToxPrint chemotypes: chain:alkaneLinear_ethyl_C2
bond:CN_amine_aromatic_generic
bond:COH_alcohol_sec-alkyl
chain:alkeneBranch_mono-ene_2-butene
chain:alkaneLinear_propyl_C3
bond:N(=O)_nitro_aromatic
bond:COH_alcohol_aliphatic_generic
bond:C=O_carbonyl_ab-unsaturated_aliphatic
chain:alkaneLinear_butyl_C4
ring:aromatic_benzene
Physicochemical properties: HOMO/LUMO logP logS topological polar surface area (TPSA) dipole moment hydrogen bond acceptors and donors complexity rotatable bonds diameter principle moments of inertia https://chemotyper.org https://toxprint.org
Chemotype knowledge
https://chemotyper.org
Importance and frequency of chemical-
protein reactivity classes in training set
Each mode-of-action (MoA) model is trained independently to optimize parameters and provide reliability estimates used in subsequent decision theory analysis.
Ordinal potency classification: non-sensitizing (1), weak (2), moderate (3), strong/extreme (4)
1 2 3 4 1 111 15 11 1 2 25 51 28 0 3 9 25 70 18 4 2 4 30 56
Global model training set results
predicted
actu
al
d+1 reliabilities = [0.16 0.20 0.24]
d0 reliabilities = [0.76 0.54 0.50 0.75]
d-1 reliabilities = [0.17 0.26 0.22]
Global model applied to test set (54 compounds)
chemotype categories physicochemical properties
skin metabolic rules
yes In domain of applicability?
{1} {1,2} {2} {2,3} {3} {3,4} {4} {1,2,3,4} 1 15 4 0 1 0 0 0 0 2 5 6 0 2 0 3 0 0 3 0 0 0 3 0 2 2 1 4 0 1 0 0 0 6 3 0
predicted
actu
al
concordance = 76% ordinal agreement (g statistic) = 0.83
Uncertainty analysis results in no prediction for only one compound.
_____________Individual model results__________ Combination of evidence (qWoE)
Molecular initiating events in sensitization: hapten-protein adduct
formation reactions:
global SN2 MoA
0,0
0,2
0,4
0,6
0,8
1,0
belief plausibility
0,0
0,2
0,4
0,6
0,8
1,0
belief plausibility
0,0
0,2
0,4
0,6
0,8
1,0
belief plausibility
Altex 29, 4/12, 373-378, Novel Technologies and an Overall Strategy to Allow Hazard Assessment and Risk Prediction of Chemicals, Cosmetics, and Drugs with Animal-Free Methods