Detection of outliers is a complex and challenging area of research in chemical theory. Among current notions, that of outliers in the chemical space-descriptors-is meaningful with multiple applications in the field of drug discovery and predictive modeling. Presented here is a new framework for outlier detection, relying on a discrete, fragment-based representation of the molecular structures. From this starting point, a recursive method is developed that quantifies the contribution of fragments to compound description and identifies outliers in chemical structure databases according to a novel definition. In contrast to existing detection routes, this approach avoids the use of thresholds usually required to quantify outlying behavior. Three chemical databases are investigated to demonstrate its generality and flexibility. The result reveals a new species of outliers, compounds with no specific structural features, rather than unique ones.
ASJC Scopus subject areas
- Chemical Engineering(all)
- Computational Theory and Mathematics
- Computer Science Applications
- Information Systems
- Library and Information Sciences