Summary
A bioRxiv preprint presents Z-Hunt-DP, a dynamic-programming reformulation of a thermodynamic Z-DNA prediction method. The authors report that it preserved the original minimum-energy result while producing a 3.30 × 10^5 speedup at 20 dinucleotide positions.
A new computational method called Z-Hunt-DP reduces the cost of predicting DNA sequences that may adopt the less common Z-DNA structure, according to a preprint posted on bioRxiv on September 20, 2026. The method keeps the thermodynamic objective used by the earlier Z-Hunt approach but replaces its exhaustive conformation search with dynamic programming.
Z-DNA is a left-handed DNA conformation that has been implicated in gene regulation and chromatin dynamics. Most cellular DNA is found in the canonical B-DNA form. Because Z-DNA is generally less thermodynamically favourable under physiological conditions, computational prediction can help identify sequences that are more likely to form it and guide analysis of experimentally observed loci.
Replacing an exponential search
The original Z-Hunt method uses a dinucleotide thermodynamic model. In a sequence window containing d dinucleotide positions, it searches possible anti/syn assignments at each position. The number of configurations therefore grows exponentially, expressed in the preprint as O(2^d).
Z-Hunt-DP reorganises the same optimisation problem so that partial solutions can be reused as the algorithm moves through the window. This changes the search cost to O(d) while retaining the original thermodynamic objective. In practical terms, the method avoids separately evaluating every complete assignment when many subproblems overlap.
On benchmark windows, the authors report that Z-Hunt-DP matched the minimum energy found by brute force within numerical tolerance. At a window containing 20 dinucleotide positions, the reported speedup was 3.30 × 10^5.
Results on mapped human loci
The authors also tested interval localisation using public human genomic loci with experimentally mapped Z-DNA. Z-Hunt-DP recovered the clipped reference interval in all 24 cases. It was also reported to be the most stable localiser in analyses using both midpoint-core and expanded-panel definitions of the target interval.
These tests address whether the predicted regions align with mapped Z-DNA intervals. They are not presented as a single, directly comparable thermodynamic ranking: the comparison set included thermodynamic, heuristic and learned models. The preprint therefore interprets the cross-tool results as localisation comparisons rather than direct tests of thermodynamic scores.
Why the algorithm matters
The main advance is computational rather than the discovery of a new DNA structure or a new biological mechanism. By reducing the conformation search from exponential to linear scaling with the number of dinucleotide positions, Z-Hunt-DP could make thermodynamic Z-DNA analysis more practical for larger or repeated sequence searches, subject to the model and benchmark conditions used in the study.
The work is currently a preprint rather than a peer-reviewed research article. The authors have made the source code and benchmark materials available through the project repository, allowing the method and reported comparisons to be examined and reproduced.