Related Experiment Videos
Predicting HIV drug resistance with neural networks
Sorin Drăghici1, R Brian Potter
1Department of Computer Science, 431 State Hall, Wayne State University, Detroit, MI 48202, USA. sod@cs.wayne.edu
Bioinformatics (Oxford, England)
|December 25, 2002
Summary
Predicting HIV protease drug resistance is crucial for effective HIV therapy. This study developed computational models using structural and sequence data, achieving up to 78% accuracy in predicting resistance to protease inhibitors.
Area of Science:
- Computational biology
- Drug discovery
- Bioinformatics
Background:
- HIV drug resistance significantly impacts treatment efficacy.
- Predicting resistance in HIV protease mutants can guide the development of improved therapies.
Purpose of the Study:
- To develop computational models for predicting HIV protease drug resistance.
- To evaluate predictors based on structural and sequence data for HIV protease inhibitors Indinavir and Saquinavir.
Main Methods:
- Utilized self-organizing maps for feature extraction and pattern clustering.
- Constructed predictors based on HIV protease-drug complex structural features (contacts).
- Developed classifiers using sequence data from drug-resistant HIV mutants.
Main Results:
- Structural-based classifier achieved 60-70% accuracy.
- Sequence-based single classifier reached 68% accuracy and 69% coverage.
- Combined multiple networks yielded 78% accuracy and 85% coverage, significantly outperforming random prediction.
Conclusions:
- Computational prediction of HIV protease drug resistance is feasible.
- Combining multiple predictive models enhances accuracy and coverage.
- These models can aid in designing more effective HIV treatment regimens.