Related Experiment Videos
Predicting nucleic acid torsion angle values using artificial neural networks
M L Beckers1, W J Melssen, L M Buydens
1Laboratory for Analytical Chemistry, Faculty of Science, University of Nijmegen, The Netherlands.
Journal of Computer-Aided Molecular Design
|May 7, 1998
Summary
This study introduces a new artificial neural network method to predict nucleic acid dinucleotide torsion angles. The model accurately predicts angles, aiding in conformational analysis and structure determination.
Area of Science:
- Computational Biology
- Structural Biology
- Bioinformatics
Background:
- Accurate prediction of nucleic acid dinucleotide torsion angles is crucial for understanding DNA and RNA structures.
- Existing methods may have limitations in speed or accuracy for conformational analysis.
Purpose of the Study:
- To develop a novel computational method for predicting key torsion angles (chi, zeta, alpha) in nucleic acid dinucleotides.
- To assess the predictive accuracy of the developed model using known crystal structure data.
Main Methods:
- An error back-propagation artificial neural network was employed.
- Training and test datasets comprising 163 and 81 nucleic acid dinucleotides with known crystal structures were utilized.
- A three-layered network with 7 hidden units was constructed.
Main Results:
- The developed artificial neural network model demonstrated good predictive ability for torsion angles.
- Approximately 70-80% of the predicted torsion angles exhibited residuals smaller than 10 degrees.
- The model's performance suggests utility in generating initial trial structures for conformational analysis.
Conclusions:
- The artificial neural network model provides a reliable approach for predicting nucleic acid dinucleotide torsion angles.
- The method facilitates the construction of preliminary molecular structures for further refinement.
- The approach is extendable for predicting torsion angles in nucleic acid structures in solution using experimental data (e.g., COSY experiments).