Home- Conferences
- Program
- Discovery & Development Forum
- Translational Bioinformatics
- Systems Biology
- Blood Brain Barrier
- Predictive Toxicology
- Berg, E
- Bhal, S
- Bryant, S
- Elkin, P
- Floudas, C
- Fraczkiewicz, R
- Hardy, B
- Judson, R
- Mu, F
- No, K
- Patlewicz, G
- Pelekis, M
- Poroikov, V
- Tsaioun, K
- Location
- Sponsors & Exhibitors
- Call for Contributions
- Bursary Awards
- Poster Session
- Photo Gallery
- Workshops & Training
- Program
- Registration
- About
- Overview
- Contact
- Support
- Schedule
|
|
|
|
|
|
|
|
|
| About Fangping Mu (Los Alamos National Laboratory) |
|
Fangping Mu, after completing an MSc in Chemical Engineering from Tsinghua University, China, finished his PhD in Chemical Engineering at Purdue University, Lafayette, Indiana. In 2005, he joined the Los Alamos National Laboratory, where he is currently a scientist in the Theoretical Biology and Biophysics Group, Theoretical Division. He has expertise in metabolomics modeling, drug
metabolism and computational chemistry. He has applied varieties of chemoinformatics and
bioinformatics tools for different metabolomics problems, such as novel metabolic reaction
identification and flux analysis. Currently, he is working on biotransformation reaction
prediction of xenobiotic compounds in mammals.
|
|
Computational Xenobiotics Metabolism Prediction System
Fangping Mu (Theoretical Biology and Biophysics, Theoretical Division, Los Alamos National Laboratory, Los Alamos, NM)
Biotransformation is the process whereby a substance, usually a foreign compound (xenobiotic),
is chemically transformed in the body to form a metabolite or a variety of metabolites. Chemical
transformations can activate a xenobiotic rendering it toxic, or can alter a xenobiotic to a nontoxic
species. The metabolism of xenobiotics is divided into phase-I and phase-II reactions,
which sometimes, but not always, occur sequentially. Xenobiotic metabolism is accomplished by
a limited number of enzymes that exhibit broad substrate specificities. These enzymes conduct
limited types of biotransformations. The majority of Phase-I reactions occur via oxidation,
including aromatic and alphatic hydroxylation, N- and O-dealkylation, deamination, and nitrogen
and sulfur oxidations. Phase-I reduction reactions include nitro reductions to amines and
dehalogenation, as well as the major hydrolysis reactions of esters and epoxides. In phase-II
glucuronidation and sulfation are the major biotransformations. Other Phase-II reactions include
acetylation, amino acid conjugation, glutathione conjugation, and methylation.
Expert systems represent the state-of-the-art xenobiotics metabolism prediction systems. These
systems are rule-based systems designed to identify functional group transformations that occur
in known reactions and then by generalizing, to formulate reaction rules for global application.
These rules can provide reasonable prediction of all possible metabolite formation. However,
they commonly predict many more metabolites than that are observed experimentally. Ranking
the possibility of metabolite formation is still not consistently available.
To overcome the significant number of false-positives in rule-based systems for metabolism
prediction, we investigated machine-learning technology for xenobiotics metabolism prediction.
We collected human xenobiotic reactions from MDL’s Metabolite database and classified
reactions according to rules based on functional group biotransformations. For each reaction rule,
the reaction center can be well defined, and is represented as a molecular substructure pattern
using SMARTs, which is a language for describing molecular patterns. Using the SMARTS
patterns, we identified potential reaction centers for each reaction class using the identified
metabolites in MDL’s Metabolite database. Each set of potential reaction centers was divided
into negative and positive examples. More than 23 atomic properties were used to model the
topological, geometrical, electronic and steric environment of the atoms in the molecule, and
more than 42 molecular properties were used to model the shape, surface, energy, and charge
distribution of the molecule. Support Vector Machines were used to separate the positive and
negative examples for each reaction class. Totally 36 biotransformations have been modeled.
Results show that the overall sensitivity and specificity of classifiers is around 87%. Prediction
of metabolism from this method can enhance the accuracy to rank the possibility of metabolite
formation.
|
|
|
|
|
|
|
|
|
|
|
