partial_order_alignment

Author: Bansho Masutani

Email: [email protected]

Short introduction

This repositry implements the Lee(2004)'s partial order alignment. Also, by adding poa_hmm = {features=["poa_simd"]} to your Cargo.toml, the POA would use SIMD instructions in some part of the alignment.

Compared to the default mode, it runs 2-3 times faster. Note that the SIMD version is entirely optional, and you should opt-in these feature and use the nightly compiler to activate this feature.

Here are some recommendations and remarks;

• The alignment is a semi-global one. In other words, it allows to leave some leading/trailing node that does not align to any query sequence.
• The POA graph is not tuned to generate consensus currently. I recommend to benchmark by some synthetic datasets.
• Unless noted, the alignment is the optimal one. It allocates and fills all the cells of the DP matrix. If it is so slow, please use banded functions with an appropriate diameter parameter. If you do not use any noisy dataset such as PacBio's or ONT's long reads, it is the case.
• There is no way to obtain 2nd or 3rd optimal alignment.
• This crate not only provides you to construct a POA graph but also enables you to calculate the likelihood of a sequence from the given POA graph. See ./src/forward.rs for more details. Currently, I use this functionality for clustering long-reads into each haplotype.

License

MIT license