Sequencing by Hybridization – A Simulation Study of Performance on Genomic Sequences

Sequencing by Hybridization (SBH)[1,2] is a theoretical method for de-novo sequencing of DNA by means of reconstruction of the sequence from its hybridization pattern. Typically, arraying a complete set of k-mers is considered, although different setups, such as degenerate probe arrays, are also possible [3,5]. While this method is not competitive with current biochemical sequencing methods, variations of the model may be practical for problems such as SNP genotyping. Formally, given a DNA sequence S of length n and a probe length k (where k<<n), hybridization of S on an array of all possible k-mers defines a hybridization pattern ) (S k σ . Given ) (S k σ , the length n, and the k-prefix and k-suffix of S, SBH algorithms attempt to reconstruct the sequence S [4]. Even when assuming perfect and stringent hybridization SBH still suffers from an inherent information problem. For any given k, there exist sequences that cannot be uniquely reconstructed by SBH. Specifically, sequences that contain a double alternating repeat of (k-1)-mers ( e d c b a β α β α ) or a triple repeat ( d c b a γ γ γ ) cannot be uniquely reconstructed. Therefore, for a given k, the probability of success of SBH in uniquely reconstructing randomly generated sequences decreases rapidly as sequence length, n, increases. Since, from the information point of view, (k-1)-mer repeats are the basis of the failure mechanism we expect performance to be even worse on genomic sequences. In this report, we describe an analysis by simulations of SBH performance on genomic sequences. We specifically test the extent to which increasing order Markov models of the data explain SBH poor performance on genomic sequences.