Fig. 4

An instance with three strings (haplotypes) of length 1000 where spectral clustering on G fails. The coverage is 10, the effective read lengths r=2 and R=250. The error probability p=0.05. The figure on the left is the union of the three figures on the right. The three colored plots are the spatial locations of the recovered communities by the spectral algorithm applied on G. The density of the recovered blue estimates is 0.0995, while that of red is 0.33 and of green is 0.57. The total overlap achieved by the spectral method is 0.1. Note that in the ground truth all three colors are equal in intensity, which is not captured by the spectral method. However, our correctly algorithm predicts approximately equal-sized communities, achieves an overlap of 0.98 and runs 4 times faster