Predicting biotransformation potential from molecular structure

J Chem Inf Comput Sci. 2003 Sep-Oct;43(5):1636-46. doi: 10.1021/ci034078l.


The program PASS-BioTransfo is presented, which is capable of predicting many classes of biotransformation for chemical compounds. A particular class of biotransformation is defined by the chemical transformation type and may additionally include the name of the enzyme involved in a transformation. An evaluation of the approach is presented, using biotransformations taken from the databases Metabolite (MDL) and Metabolism (Accelrys), respectively. When trained with biotransformations from Metabolite, PASS-BioTransfo predicts 1927 classes of biotransformation; the average accuracy estimated in LOO cross-validation is about 88%. After training with the biotransformations from the Metabolism database, 178 classes of biotransformation are predicted with an average accuracy of about 85%. The results of cross-prediction with several training and evaluation sets are presented and discussed.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Animals
  • Biotransformation*
  • Databases, Factual
  • Humans
  • Molecular Structure
  • Pharmaceutical Preparations / chemistry*
  • Pharmaceutical Preparations / metabolism
  • Software*
  • Structure-Activity Relationship


  • Pharmaceutical Preparations