Related Experiment Videos
Implementation of the three-dimensional-pattern search problem on Hopfield-like neural networks
E Feuilleaubois1, V Fabart, J P Doucet
1Institut de Topologie et de Dynamique des Systèmes (ITODYS), CNRS UA-34, Université Paris VII, France.
SAR and QSAR in Environmental Research
|January 1, 1993
Summary
This study introduces a novel Hopfield-like neural network approach for the 3D-pattern search problem in molecules. The method efficiently identifies similar spatial arrangements and offers flexibility for conformational analysis.
Area of Science:
- Computational Chemistry
- Bioinformatics
- Artificial Intelligence
Background:
- The 3D-pattern search problem involves identifying atom subsets with spatial arrangements similar to a given 3D pattern within a molecule.
- This is a computationally challenging NP-complete combinatorial optimization problem.
Purpose of the Study:
- To develop a new method for solving the 3D-pattern search problem using Hopfield-like neural networks.
- To leverage neural networks for combinatorial optimization in molecular pattern recognition.
Main Methods:
- An objective function was formulated based on the differences in interatomic distances between the pattern and the molecule.
- This objective function was implemented and optimized using Hopfield-like neural networks.
Main Results:
- The proposed method successfully retrieves the specified 3D pattern within a molecule.
- The approach can also identify partial solutions, suggesting subsets with fewer atoms, which is useful for analyzing local conformational flexibility.
Conclusions:
- Hopfield-like neural networks provide an effective platform for combinatorial optimization in 3D-pattern searching.
- The distributed representation on these networks shows promise for parallel implementation, enhancing computational efficiency.