[Paper Review] Optimal DNA shotgun sequencing: Noisy reads are as good as noiseless reads

Seyed Abolfazl Motahari, Kannan Ramchandran+2 more|arXiv (Cornell University)|Apr 9, 2013

Algorithms and Data Compression7 references4 citations

TL;DR

This paper establishes that for i.i.d. DNA sequences, noisy shotgun reads are as effective as noiseless reads for sequence reconstruction, provided the noise level is below a critical threshold—up to 19% for symmetric substitution noise. The key insight is that high-coverage sequencing enables error correction via alignment and averaging, making noise asymptotically irrelevant when read length exceeds the longest repeat in the genome.

ABSTRACT

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%.

Motivation & Objective

• To determine the fundamental limits of DNA shotgun sequencing under noisy reads.
• To investigate whether noise fundamentally degrades reconstruction performance in the asymptotic regime.
• To identify conditions under which noisy reads can achieve the same performance as noiseless reads.
• To analyze the role of coverage depth and read length in enabling error correction despite noise.
• To establish a theoretical threshold on noise beyond which performance degrades, independent of read length.

Proposed method

• Proposes a two-stage assembly scheme: first, align multiple noisy reads covering the same genomic region using a maximum-likelihood alignment criterion.
• Applies a cleaning procedure that averages aligned reads to produce a consensus sequence, reducing noise through redundancy.
• Uses large deviation bounds to analyze the probability of misalignment and incorrect consensus estimation.
• Derives a critical threshold condition I_read > H_renyi, where I_read is the mutual information between true base and observed read, and H_renyi is the Rényi entropy rate of order 2.
• Employs coverage analysis based on Poisson-distributed read counts and union bounds to show that full genome coverage is achievable under the threshold condition.
• Applies concentration inequalities and exponential bounds to prove that the probability of reconstruction error vanishes as genome size G → ∞.

Experimental results

Research questions

• RQ1 Under what conditions can noisy reads achieve the same reconstruction performance as noiseless reads in shotgun sequencing?
• RQ2 What is the fundamental threshold on noise level below which noise has no asymptotic impact on reconstruction?
• RQ3 How does high-coverage sequencing enable error correction in the presence of noisy reads?
• RQ4 Is the critical read length required for reconstruction under noise the same as under noiseless conditions?
• RQ5 Can the coverage bound (Lander-Waterman) be asymptotically tight even with noisy reads?

Key findings

• For i.i.d. DNA sequences, noisy reads are as effective as noiseless reads if the noise level is below a threshold, specifically when I_read > H_renyi.
• For a uniformly distributed DNA sequence and symmetric substitution noise, the noise threshold is δ* = 0.19, meaning up to 19% misread probability causes no asymptotic performance loss.
• The critical read length required for reconstruction remains L > 2 / H_renyi, the same as in the noiseless case, indicating that noise does not increase the minimum required read length.
• High-coverage sequencing (coverage depth → ∞ as G → ∞) enables effective error correction via alignment and averaging of multiple noisy reads from the same genomic region.
• The probability of reconstruction error, including misalignment and incorrect consensus, vanishes as G → ∞ when the noise threshold condition is satisfied and coverage is sufficient.
• The coverage condition N > (G / L) ln(G) ensures that all bases are covered with high probability, and this condition is sufficient for reconstruction under noisy reads when the noise is below the threshold.