A new computational method for detection of chimeric 16S rRNA artifacts generated by PCR amplification from mixed bacterial populations

A new computational method (chimeric alignment) has been developed to detect chimeric 16S rRNA artifacts generated during PCR amplification from mixed bacterial populations. In contrast to other nearest-neighbor methods (e.g., CHECK_CHIMERA) that define sequence similarity by k-tuple matching, the chimeric alignment method uses the score from dynamic programming alignments. Further, the chimeric alignments are displayed to the user to assist in sequence classification. The distribution of improvement scores for 500 authentic, nonchimeric sequences and 300 artificial chimeras (constructed from authentic sequences) was used to study the sensitivity and accuracy of both chimeric alignment and CHECK_CHIMERA. At a constant rate of authentic sequence misclassification (5%), chimeric alignment incorrectly classified 13% of the artificial chimeras versus 14% for CHECK_CHIMERA. Interestingly, only 1% of nonchimeras and 10% of chimeras were misclassified by both programs, suggesting that optimum performance is obtained by using the two methods to assign sequences to three classes: high-probability nonchimeras, high-probability chimeras, and sequences that need further study by other means. This study suggests that k-tuple-based matching methods are more sensitive than alignment-based methods when there is significant parental sequence similarity, while the opposite becomes true as the sequences become more distantly related. The software and a World Wide Web-based server are available at http://www-hto.usc.edu/software/mglobalCHI.