Acute toxicity (LD50) modeling utilizing fragmental QSAR, similarity analysis and reliability of prediciton

TOXI 47

Pranas Japertas, jurgutis@ap-algorithms.com1, Remigijus Didziapetris2, and Alanas Petrauskas1. (1) Pharma Algorithms Inc, 591 Indian Rd., Toronto, ON M6P 2C4, Canada, (2) Pharma Algorithms, Inc, 591 Indian Road, Toronto, ON, Canada
Acute toxicity (LD50) can be viewed as ‘cumulative potential' to cause death in animals. Complex endpoints such as acute toxicity involve a variety of mechanisms. Reliable data sets are limited and the investigated chemical space is small. Here we present an LD50 model with estimation of reliability of prediction. Data for over 100,000 compounds with LD50 for six animal systems - Rat oral, intraperitoneal, Mouse oral, intraperitoneal, intravenous, subcutaneous - were compiled and critically analyzed. Fragmental QSAR was used to build a baseline prediction model and corrected through analysis with a novel structural similarity key. A reliability index considers similarity of test compound to training set, distance between predicted LD50 and experimental values, consistence of experimental values for similar compounds. Prediction results with high reliability index are accurate to rmse < 0.5 log unit. An acceptable reliability index usually is reported for 30-50% of compounds in different external test sets.
 

Poster Session and Awards
6:00 PM-10:00 PM, Tuesday, August 21, 2007 BCEC -- 204 A/B, Poster

Division of Chemical Toxicology

The 234th ACS National Meeting, Boston, MA, August 19-23, 2007