The application of Markov chain analysis to oligonucleotide frequency prediction and physical mapping of Drosophila melanogaster

Here we compare several methods for predicting oligonucleotide frequencies in 691 kb of Drosophila melanogaster DNA. As in previous work on Escherichia coli and Saccharomyces cerevisiae, a relatively simple equation based on tetranucleotide frequencies can be used in predicting frequencies of higher order oligonucleotides. For example, the mean of observed/expected abundances of 4,096 hexamers was 1.07 with a sample standard deviation of .55. This simple predictor arises by considering each base on the sense strand of D. melanogaster to depend only on the three bases 5' to it (a 3rd order Markov chain) and is more accurate than the random predictor. This equation is useful in predicting restriction enzyme fragment sizes, selecting restriction enzymes that cut preferentially in coding vs noncoding regions, and in selecting probes to fingerprint clones in contig mapping. Once again, this equation well predicts the occurrence of higher order oligonucleotides, supporting our hypothesis that this predictor holds in evolutionarily diverse organisms. When ranked from highest to lowest abundance, the observed frequencies of oligomers of a given length are closely tracked by the predicted abundances of a 3rd order Markov chain. Through use of the dependence of oligomer frequencies on base composition, we report a list of oligomers that will be useful for the completion of a cosmid physical map of D. melanogaster. Presently, the library is such that it will be possible to construct large contigs using only 30 oligonucleotide probes to fingerprint cosmids.