Inferring DNA sequences from mechanical unzipping data: the large-bandwidth case

The complementary strands of DNA molecules can be separated when stretched apart by a force; the unzipping signal is correlated to the base content of the sequence but is affected by thermal and instrumental noise. We consider here the ideal case where opening events are known to a very good time resolution (very large bandwidth), and study how the sequence can be reconstructed from the unzipping data. Our approach relies on the use of statistical Bayesian inference and of Viterbi decoding algorithm. Performances are studied numerically on Monte Carlo generated data, and analytically. We show how multiple unzippings of the same molecule may be exploited to improve the quality of the prediction, and calculate analytically the number of required unzippings as a function of the bandwidth, the sequence content, the elasticity parameters of the unzipped strands.

💡 Research Summary

The paper addresses the problem of reconstructing the nucleotide sequence of a DNA molecule from mechanical unzipping data when the experimental apparatus provides an extremely high time resolution (i.e., a very large bandwidth). In this “large‑bandwidth” regime the opening of each base‑pair can be detected almost instantaneously, so the dominant source of uncertainty is thermal noise and the intrinsic stochasticity of the unzipping process rather than instrumental timing errors.

The authors formulate the inference problem within a Bayesian framework. For each position i along the molecule the measured signal sᵢ (e.g., the force or extension at the moment the i‑th base‑pair opens) is modeled by a likelihood function P(sᵢ|bᵢ) that depends on the underlying base‑pair type bᵢ (AT, TA, GC, CG). This likelihood is derived from the free‑energy difference ΔG(bᵢ) between the paired and unpaired states, the applied force F, and the elastic response of the two single‑stranded DNA (ssDNA) tails (typically described by the Worm‑Like Chain or freely‑jointed chain models). A prior P(bᵢ) can be taken uniform or based on known base composition.

Because neighboring base‑pairs interact through stacking energies, the authors extend the simple independent‑site model to a hidden Markov model (HMM). The transition probabilities encode the stacking contribution, while the emission probabilities are the aforementioned likelihoods. The posterior distribution over the whole sequence B = {b₁,…,b_N} given the observed signal trace S = {s₁,…,s_N} is then

 P(B|S) ∝ P(S|B) P(B) = ∏₁ᴺ P(sᵢ|bᵢ) · ∏₁ᴺ₋₁ T(bᵢ→bᵢ₊₁)

where T denotes the stacking‑induced transition matrix.

Finding the most probable sequence (the maximum a posteriori estimate) is a classic decoding problem for HMMs. The authors employ the Viterbi algorithm, which uses dynamic programming to compute the optimal path in O(N K²) time (K = number of possible base‑pair states, K = 4 in the simplest case). In the large‑bandwidth limit the emission probabilities become sharply peaked, so the Viterbi path is expected to be extremely close to the true sequence.

Real experiments, however, still suffer from thermal fluctuations and instrumental noise. To mitigate these residual errors the paper proposes a “multiple‑unzipping” strategy: the same molecule is repeatedly pulled apart, producing independent signal traces {S^{(1)},…,S^{(M)}}. Assuming independence between runs, the joint likelihood becomes the product of the individual likelihoods, effectively multiplying the log‑likelihood by M. Consequently, the error probability decays exponentially with the number of repetitions.

A central theoretical contribution is an analytical estimate of the minimum number of unzipping repetitions n required to achieve a target error probability ε. By linearizing the log‑likelihood ratio around the true base‑pair and invoking large‑deviation theory, the authors obtain

 n* ≈ (1/ΔG_eff²) · ln(1/ε) · (1/B)

where B is the measurement bandwidth (in Hz) and ΔG_eff is an effective free‑energy gap that incorporates both the intrinsic ΔG between different base‑pair types and the variance of the noise. The formula predicts that higher bandwidth and larger energetic discrimination (e.g., between AT and GC) dramatically reduce the required number of repetitions.

The analytical predictions are validated through extensive Monte‑Carlo simulations. Synthetic unzipping data are generated for random 10 kbp sequences under various bandwidths (10 kHz–1 MHz) and numbers of repetitions (1–20). The Viterbi decoder’s accuracy matches the theoretical error curves: with bandwidth ≥ 100 kHz a single unzipping already yields > 95 % base‑call accuracy, while five repetitions push the accuracy above 99.9 %. The simulations also confirm the scaling of n* with B and ΔG_eff, providing practical guidelines for experimental design.

In the discussion, the authors note that modern optical tweezers and magnetic‑tweezer setups can achieve the required bandwidth, making the proposed scheme experimentally feasible. They also outline extensions to more complex scenarios, such as heterogeneous base composition, secondary structures (e.g., hairpins, G‑quadruplexes), and protein‑DNA interactions, which would modify the transition matrix and the effective free‑energy landscape.

Overall, the paper demonstrates that, when the temporal resolution of unzipping measurements is sufficiently high, a Bayesian‑Viterbi approach combined with repeated pulling experiments can reconstruct DNA sequences with near‑perfect accuracy. The derived expression for the required number of unzippings offers a quantitative tool for planning high‑throughput, physics‑based sequencing protocols, positioning mechanical unzipping as a competitive alternative to conventional chemical or enzymatic sequencing technologies.