The complexity of string partitioning

Given a string $w$ over a finite alphabet $\Sigma$ and an integer $K$, can $w$ be partitioned into strings of length at most $K$, such that there are no \emph{collisions}? We refer to this question as the \emph{string partition} problem and show it is \NP-complete for various definitions of collision and for a number of interesting restrictions including $|\Sigma|=2$. This establishes the hardness of an important problem in contemporary synthetic biology, namely, oligo design for gene synthesis.

💡 Research Summary

The paper introduces a decision problem that we call String Partition: given a finite alphabet Σ, a word w∈Σ*, and an integer K, can w be cut into substrings (called pieces) each of length at most K such that no two pieces “collide”? The notion of collision is deliberately left flexible, and the authors explore several natural definitions that arise in the context of synthetic biology, especially oligonucleotide design for gene synthesis.

Collision definitions. Four principal types are examined:

1. Overlap collision – two pieces share at least one position in the original string. This models the physical impossibility of using overlapping oligos in a synthesis protocol.
2. Substring collision – one piece appears as a contiguous substring of another piece. In a laboratory setting this would cause a longer oligo to bind nonspecifically to a shorter one that is fully contained within it.
3. Identity collision – two distinct pieces are exactly the same string (or have the same hash/biophysical signature). Identical oligos increase the risk of cross‑hybridisation and waste reagents.
4. Generalised collision – any user‑specified binary relation on pieces (e.g., equal GC‑content, presence of a particular motif, or a custom similarity metric).

For each definition the decision problem asks whether a collision‑free K‑bounded partition exists.

NP‑completeness proof strategy. The authors reduce the classic NP‑complete problem 3‑SAT to String Partition. For a given 3‑CNF formula they construct a word w that encodes variables, literals, and clauses as carefully designed blocks of symbols. The key idea is to force any feasible partition to choose, for each variable, either a “true” block or a “false” block, but not both, because choosing both would create a forbidden collision under the chosen definition. Clause blocks are built so that at least one of their three literal blocks must be compatible with the variable choices; otherwise a collision becomes unavoidable. The integer K is set large enough to accommodate each block as a single piece but small enough that splitting a block would break the intended correspondence. The reduction runs in polynomial time and preserves satisfiability: the formula is satisfiable iff the constructed instance admits a collision‑free partition.

Binary alphabet restriction. While many reductions assume an unrestricted alphabet, the paper shows that the hardness persists even when |Σ|=2. This is achieved by encoding each original symbol as a fixed‑length binary codeword and inserting delimiter symbols that separate codewords without creating new collisions. The binary encoding preserves the structure of the original reduction, so the NP‑completeness carries over to the binary case. This result is particularly relevant because DNA sequences are naturally over a four‑letter alphabet, and any practical synthesis tool must work with a constant‑size alphabet.

Implications for synthetic biology. In gene synthesis, a designer must select a set of oligonucleotides (typically 20–60 bases long) that collectively cover a target DNA sequence without overlaps, inclusions, or other undesirable interactions. The String Partition problem captures exactly this design step: each piece corresponds to an oligo, the bound K reflects the maximum oligo length, and collisions model the biochemical constraints that would cause synthesis failure or off‑target binding. By proving that even the simplest version of the problem (binary alphabet, overlap‑only collisions) is NP‑complete, the authors demonstrate that no polynomial‑time algorithm can guarantee an optimal oligo set unless P=NP. Consequently, existing software tools rely on heuristics, greedy strategies, or integer‑linear‑programming formulations that may produce suboptimal solutions.

Special cases and algorithmic outlook. The paper also discusses tractable subclasses. If K is a fixed constant and the collision relation is limited to simple overlaps, dynamic programming can solve the problem in O(n·K) time, where n=|w|. Similarly, when the alphabet size is unbounded but the collision relation is restricted to exact identity, the problem reduces to a variant of the classic “minimum segmentation” problem, which is solvable in polynomial time. However, these special cases do not capture the full range of constraints encountered in real‑world oligo design, where multiple collision types coexist and K is dictated by experimental chemistry rather than by algorithmic convenience.

Future research directions. The authors suggest several avenues: (i) extending the hardness proof to more biologically realistic collision models that incorporate thermodynamic stability or secondary‑structure formation; (ii) investigating parameterised complexity with respect to K, the number of distinct collision types, or the maximum degree of the collision graph; (iii) developing approximation algorithms with provable guarantees for the general case; and (iv) empirical studies of average‑case hardness on randomly generated DNA sequences to assess how often the worst‑case behaviour manifests in practice.

In summary, the paper provides a rigorous computational‑complexity foundation for a central problem in synthetic biology. By formalising string partitioning, exploring multiple collision notions, and establishing NP‑completeness even under severe alphabet restrictions, it clarifies why oligo design remains a challenging combinatorial task and motivates the continued development of sophisticated heuristic and approximation methods.