Optimal DNA Shotgun Sequencing: Noisy Reads are as Good as Noiseless Reads

Optimal DNA Shotgun Sequencing: Noisy Reads are as Good as Noiseless Reads

This arxiv paper is making the rounds in twitter today.

We establish the fundamental limits of DNA shotgun sequencing under noisy reads. We show a surprising result: for the i.i.d. DNA model, noisy reads are as good as noiseless reads, provided that the noise level is below a certain threshold which can be surprisingly high. As an example, for a uniformly distributed DNA sequence and a symmetric substitution noisy read channel, the threshold is as high as 19%.

It uses concepts from information theories to define the limitations of sequencing technologies. We covered some of the basic concepts in a blog post more than a year back. Basic idea is this. Bioinformaticians tend to add too much premium to perfect sequencing than slightly imperfect sequencing, but that premium could be due to limitations of existing algorithms than the sequencing technology. In digital and analog communication, data gets transferred from one point to another, even though the channels have some degree of noise. In fact, when signals are transmitted from Mars to Earth, the channel is extremely noise, yet the the original transmitted images can be reconstructed with high degree of accuracy.

The above Berkeley group has been doing very interesting mathematical work on sequencing, and we covered their earlier paper few months back. Many readers made insightful comments on Optimal Assembly for High Throughput Shotgun Sequencing

Going through the current paper, here are the key concepts. In noiseless systems,

The optimal assembly algorithm which achieves the fundamental limit in the above setting is the greedy algorithm.

What happens in noisy sequencer?

The modi?cation of the greedy algorithm is only one approach to deal with noise. But are there better approaches? What, in fact, is the fundamental limit on the system performance under noisy reads? We show a surprising result in this paper: provided that the noise level is below a certain threshold, noise has no impact on the asymptotic performance

What it means for assembly.

The scheme we propose to achieve the fundamental limit under noisy reads has two stages: an error-correction phase, which aligns reads from the same region of the DNA and averages across them to produce cleaner reads, followed by an assembly phase, applying the greedy algorithm with approximate match to the cleaner reads. Provided that the noise level satis?es condition (1) to allow accurate read alignment, the noise level of the reads can be driven to be vanishingly small after the error-correction phase. Since it was shown in [7] that the performance of the greedy algorithm is continuous in the noise level, this implies noiseless performance can be achieved asymptotically.

In the assembly literature, there are two approaches to deal with the noise in the reads. In the ?rst approach, error-correction is performed jointly with assembly such as Velvet [10] and ABySS [4] which are based on de Bruijn graph. In the second approach, error correction is performed ?rst, followed by an assembly algorithm which assumes the reads are essentially clean. Examples of the algorithms are SHREC [3], Reptile [9], and Quake [5]. The latter is a separation approach, which is conceptually simpler. What we show in this paper is that, at least for the simple model considered here, the separation approach is in fact information-theoretically optimal, up to a certain threshold on the noise level.

Bottom line - just like we saw in case of internet search, once again electrical engineers win over computer scientists :)


Our friend Dr. Chen-Shan Chin provides some insight on the next difficult problem in assembly -

Our advance (at both intellectual level and practical level) is more on new algorithms to handle indel errors better. (In communication theory, this is related to “insertion-deletion channel” problem. There are still a lot of open questions about it.)

Written by M. //