Discovery of frequent DATALOG patterns
Data Mining and Knowledge Discovery
Multi-task learning for HIV therapy screening
Proceedings of the 25th international conference on Machine learning
PKDD'05 Proceedings of the 9th European conference on Principles and Practice of Knowledge Discovery in Databases
Relational Feature Mining with Hierarchical Multitask kFOIL
Fundamenta Informaticae - Machine Learning in Bioinformatics
Hi-index | 0.00 |
HIV therapy optimization is a hard task due to rapidly evolving mutations leading to drug resistance. Over the past five years, several machine learning approaches have been developed for decision support, mostly to predict therapy failure from the genotypic sequence of viral proteins and additional factors. In this paper, we define a relational representation for an important part of the data, namely the sequences of a viral protein (reverse transcriptase), their mutations, and the drug resistance( s) associated with those mutations. The data were retrieved from the Los Alamos National Laboratories' (LANL) HIV databases. In contrast to existing work in this area, we do not aim directly for predictive modeling, but take one step back and apply descriptive mining methods to develop a better understanding of the correlations and associations between mutations and resistances. In our particular application, we use the Warmr algorithm to detect non-trivial patterns connecting mutations and resistances. Our findings suggest that well-known facts can be rediscovered, but also hint at the potential of discovering yet unknown associations.