DNA-Inspired Information Concealing

Protection of the sensitive content is crucial for extensive information sharing. We present a technique of information concealing, based on introduction and maintenance of families of repeats. Repeats in DNA constitute a basic obstacle for its reconstruction by hybridization.

💡 Research Summary

The paper introduces a novel information‑concealing technique inspired by the abundance of repeat families in eukaryotic DNA. The authors observe that repeated DNA segments make it difficult to reconstruct the original genome by hybridisation, and they translate this biological principle into a computational method for protecting sensitive data while still allowing the sharing of short, locally useful substrings.

The core problem is formally defined as follows: given an input sequence ω and a small integer k, produce a transformed sequence ω′ such that (I) every substring of ω of length ≤ k also appears in ω′, (II) reconstructing ω from ω′ is computationally hard, (III) the length of ω′ is linear in |ω|, and (IV) with low probability ω′ contains substrings that never occurred in ω, while preserving the frequency rank of the retained substrings. This formulation captures the dual goals of preserving enough local information for tasks like worm detection or intrusion analysis, yet preventing an adversary from recovering the full original content.

The proposed algorithm revolves around a basic building block called Procedure S. The input is first turned into a cyclic string, then partitioned into disjoint blocks whose lengths are randomly chosen between lower and upper bounds (lb, ub). For each block an overlap of fixed size o is prepended; this overlap guarantees that any k‑length substring straddling a block boundary is still present in the transformed data. Optionally, a random “dust” fragment can be appended to each block, creating an enriched unit called a “card”. All cards are finally shuffled and concatenated to form ω′. The overlap and dust together generate a large family of repeated patterns, while the random shuffling destroys the global ordering of the original sequence.

To argue that reconstruction is hard, the authors map the construction onto a de Bruijn graph: each distinct block prefix and suffix becomes a vertex, each block itself an edge from its prefix to its suffix. Because many blocks share the same prefixes and suffixes, the graph contains numerous Eulerian trails, each corresponding to a possible reconstruction of the original sequence. The number of such trails grows exponentially with the number of duplicated vertices, making the inverse problem at least as hard as finding an Eulerian trail in a highly ambiguous graph—a problem known to be NP‑hard in the general case. The paper further defines an “attacker problem” where the adversary knows ω′, the parameters (|ω|, k, o, lb, ub) and the algorithm itself, and may also have access to the frequency distribution of substrings in ω′. The authors argue that for realistic inputs (e.g., DNA, program code, multimedia streams) the attacker’s knowledge does not suffice to uniquely determine ω, especially after multiple runs of Procedure S that further scramble local order.

The related‑work section surveys existing privacy‑preserving techniques: network‑level anonymization (CryptoPan), private matching, data masking, steganography, and various data‑publishing anonymization schemes. While these methods either hide the data origin, enable specific secure computations, or embed hidden payloads, none simultaneously guarantee preservation of short‑substring statistics and provable hardness of full reconstruction. The proposed DNA‑inspired approach fills this gap by deliberately engineering repeat families that act as “self‑concealing” structures.

Complexity analysis shows that the algorithm runs in linear time and space with respect to the input size; the parameters o, lb, and ub can be tuned to balance security (more overlap and dust increase ambiguity) against overhead (larger ω′). Although the paper does not present empirical benchmarks, the theoretical results suggest that the method is practical for large‑scale data streams where preserving local patterns is essential.

In conclusion, the authors contribute a formal problem definition, a concrete construction based on overlapping blocks and random shuffling, and a hardness proof grounded in graph‑theoretic arguments. They propose future work on optimizing parameter selection, extending the model to multi‑party settings, and evaluating the scheme on real network traffic and genomic datasets. This DNA‑inspired information concealing framework offers a promising new direction for privacy‑preserving data sharing where both utility and security are required.