Pattern Recognition of Genomic Features with Microarrays: Site Typing of Mycobacterium tuberculosis Strains

Proceedings of the Twentieth International Conference on Machine Learning, 2000

Volume

Proceedings of the Twentieth International Conference on Machine Learning, 2000

Contents

Mycobacterium tuberculosis (M. tb.) strains differ in the number and locations of a transposon-like insertion sequence known as IS6110. Accurate detection of this sequence can be used as a fingerprint for individual strains, but can be difficult because of noisy data. In this paper, we propose a non-parametric discriminant analysis method for predicting the locations of the IS6110 sequence from microarray data. Polymerase chain reaction extension products generated from primers specific for the insertion sequence are hybridized to a microarray containing targets corresponding to each open reading frame in M. tb. To test for insertion sites, we use microarray intensity values extracted from small windows of contiguous open reading frames. Rank-transformation of spot intensities and first-order differences in local windows provide enough information to reliably determine the presence of an insertion sequence. The non-parametric approach outperforms all other methods tested in this study.