A semi-supervised style method for haplotype assembly problem based on MEC model

Haplotype reconstruction based on aligned SNP fragments is to infer a pair of haplotypes from localized polymorphism data got through short genome fragments. For this problem, the minimum error correction (MEC) model is one of important computational models. This model constructs a pair of haplotypes by correcting minimum SNPs in genome fragments of an individual's DNA. In this paper, a semi-supervised competitive neural network on the MEC model is proposed. This algorithm aims at clustering all fragments into two sets. The fragments in each set can then be used to construct a haplotype with minimum SNPs corrected. Although the architecture of the proposed method is simple, it outperforms other two algorithms on most instances of both real data and simulation data. So, the results show that the proposed semi-supervised neutral network is effective. The results also show that semi-supervised algorithm is feasible and promising for this problem.