Database mining using soft computing techniques. An integrated neural network-fuzzy logic-genetic algorithm approach

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Informatics

Background:

• Transition metal imido complexes (LnMdNZ) exhibit complex structural relationships.
• Understanding these relationships is crucial for predicting chemical properties and reactivity.

Purpose of the Study:

• To analyze metric parameters in transition metal imido complexes using soft computing (SC) techniques.
• To identify and quantify relationships between bond lengths and angles within these complexes.
• To explore potential novel chemical insights and database outliers.

Main Methods:

• Employed two SC techniques: competitive learning neural network and a neural network-fuzzy logic-genetic algorithm approach.
• Analyzed a subset of data from the Cambridge Structural Database.
• Utilized a fuzzy inference system to construct a multi-variable hypersurface.

Main Results:

• Confirmed and quantified a suspected inverse correlation between metal-nitrogen bond length and metal-nitrogen-carbon bond angle.
• Discovered an unexpected inverse correlation between the nitrogen-carbon (NC) distance and the metal-nitrogen-carbon (MNC) angle.
• Identified an 'outlier' cluster, potentially indicating experimental errors or novel chemical structures.
• Observed that alkyl versus aryl substituents influence these relationships.

Conclusions:

• SC techniques effectively mine chemical databases to confirm known and discover unknown structural correlations.
• The identified hypersurface highlights complex interdependencies between NC, MNred, and MNC parameters.
• The detection of outliers underscores the value of SC in identifying data anomalies and potential new chemical entities.