Information Rate Decomposition for Noisy Nanopore Channels with Geometric Duplication

Technical Summary: Information Rate Decomposition for Noisy Nanopore Channels with Geometric Duplication

Problem Statement
The paper addresses the theoretical challenge of characterizing the achievable information rates of noisy duplication channels with memory, specifically motivated by the Oxford Nanopore Technologies (ONT) sequencing process. In this physical system, the measured current signal is distorted by two primary factors:

1. Inter-Symbol Interference (ISI): Caused by multiple DNA bases residing simultaneously within the nanopore, creating a state-dependent memory effect.
2. Random Sample Duplications: Caused by variable translocation speeds, where a single channel state generates a random number of output samples.

These combined effects create a "noisy geometric duplication channel" (NNC) where the output is a noisy, piecewise-constant signal with unknown segment boundaries. Direct theoretical analysis of the mutual information rate for such channels is difficult because the distortions (memory and synchronization uncertainty) interact complexly, and the output length is random and dependent on the input.

Methodology
The authors propose a novel decomposition of the information rate that separates the channel's intrinsic memory from the uncertainty introduced by random duplications. The methodology relies on the following technical components:

• Channel Modeling: The channel is modeled using a Markov source on a de Bruijn state space (representing the nanopore state) where each state is duplicated a geometrically distributed number of times. The resulting signal is corrupted by Additive White Gaussian Noise (AWGN).
• Soft Alignment Functionals: The paper introduces a "soft alignment" functional, closely related to the Soft-Dynamic Time Warping (Soft-DTW) loss. This functional replaces the hard minimum cost of alignment paths with a "soft minimum" (a log-sum-exp over all admissible paths), weighted by a temperature parameter related to the noise level.
• Strong Asymptotic Equipartition Property (AEP): The authors establish a strong AEP for the normalized soft alignment functional. By proving the shift-subadditivity of these functionals, they apply Kingman's subadditive ergodic theorem to show that the alignment cost converges almost surely to a deterministic rate.
• Decomposition Theorem: The core analytical step involves deriving an identity that expresses the mutual information rate as the difference between an auxiliary ISI channel rate and a segmentation penalty term.

Key Contributions

1. Information Rate Decomposition: The paper derives a new decomposition formula:
   $$I_T(S; Y) = \mu(I_{\text{ISI}} - R_{\text{seg}})$$
   where $I_{\text{ISI}}$ is the achievable rate of an auxiliary ISI channel (capturing the memory of the nanopore state), and $R_{\text{seg}}$ is a penalty term capturing the uncertainty in segment boundaries caused by duplications. This separates the dominant channel distortions into two interpretable components.
2. Strong AEP for Soft Alignment: The authors prove a strong AEP-type result for the soft alignment functional. This establishes strong information stability for noisy duplication channels with ergodic inputs and provides an alternative proof of the Markov-constrained coding theorem, upgrading previous results that relied on convergence in probability to almost sure convergence.
3. Computational Tractability: The decomposition enables a more tractable Monte Carlo method for estimating information rates. The auxiliary ISI term can be evaluated using the standard forward algorithm, while the segmentation penalty is computed via a Soft-DTW-type dynamic program. This avoids the computational burden of marginalizing over both state evolution and duplication events simultaneously.
4. Jump-Reliability Bound: The authors derive an upper bound on the soft alignment penalty based on the distribution of jump distances between adjacent nanopore levels. This yields a lower bound on the information rate (the "jump-reliability bound"). This bound provides a geometric explanation of channel synchronizability: larger, more distinguishable jumps between levels reduce synchronization uncertainty and improve achievable rates.

Results and Significance
The paper claims that this framework connects information-theoretic performance directly to the geometry of the pore model and translocation dynamics. The primary significance lies in providing a tractable framework for computing achievable rates for modern nanopore sequencers (e.g., ONT R10.4.1), where the large state space makes direct estimation computationally prohibitive.

The "jump-reliability bound" formalizes the empirical intuition that distinguishable level transitions improve system performance. The authors apply this bound to estimate achievable rates for various ONT sequencers released over time, enabling comparisons that were not feasible with previous techniques. The work validates the theoretical framework by showing it recovers known results for the noisy nanopore channel under geometric duplication while offering a refined analytical tool for performance evaluation in high-sampling-rate or low-noise regimes.