Most quantum computing case studies start with a familiar problem and ask whether a quantum processor can run it faster. The project described here did something less common: it changed what the problem was.

Venturus, a Brazilian R&D institute, set out to work on food security using quantum methods, a goal framed around the UN's zero hunger objective. The path there ran through a computational bottleneck in crop breeding, and the team's answer was not a quantum port of an existing alignment algorithm. It was a new alignment algorithm, built on a reframing of what a genome sequence actually is.

This post draws on a talk by Daniel Moraes of Venturus at the QuEra Quantum Alliance 2026 Annual Meeting (QQA 2026), where he presented this work alongside two other Venturus quantum projects. We focus here on the genomics case, a collaboration between Venturus, Inari, the Open Quantum Institute (OQI), and QuEra.

Who Is Involved

Venturus is an R&D institute based in Campinas, near São Paulo, with more than 30 years of history and roughly 900 employees, plus a branch in Manaus and a sales office in Atlanta. Its project portfolio spans health, industry, mobility, utilities, telco, and agribusiness, organized around sustainable development goals. The quantum computing group inside it is deliberately small, in the range of 20 to 25 people, and it works across cryptography, enterprise quantum readiness, simulation tooling, and applied research like this one.

Inari is a Boston-based startup working on seed DNA editing. Its business depends on identifying where in a crop genome an edit should be made in order to produce more nutritious and more resilient varieties. That identification step is where the computational wall shows up.

The Open Quantum Institute (OQI) works to accelerate quantum computing applications that serve the UN Sustainable Development Goals. Its involvement was central to getting this project started.

The Scale Problem in Crop Genome Editing

Finding an edit target means aligning genomes: comparing a target sequence against a reference to determine where it corresponds. The standard, highest-accuracy methods for this are dynamic programming approaches with roughly quadratic complexity in sequence length, and quadratic behavior becomes untenable quickly at biological scale.

Venturus put concrete numbers on it. Aligning a segment of about 100 kilobases can require on the order of 40GB of memory. A tomato genome is roughly 900 megabases, several orders of magnitude larger than that already-expensive segment, and a breeding program needs to run this comparison across many genomes rather than once. The cost does not scale to the work.

A Genome as a Time Series

The conventional framing treats a genome as a static string of bases (A, C, G, T) to be pattern matched against another string. That framing is what makes the problem quadratic: every position in one sequence is, in principle, a candidate match for every position in the other.

Venturus's team stepped back from that representation. Rather than a string laid out in space, they treated the sequence of bases as a temporal series, an ordered signal unfolding over time. That single change moves the problem into a class of methods built for sequential data.

From there, the approach follows: use a quantum reservoir to train a neural network on the reference genome, then present the trained network with the target sequence and ask it where that sequence aligns. Quantum reservoir computing is well suited to this because the reservoir itself is not trained. A fixed quantum system transforms the input into a high-dimensional representation, and only a lightweight classical readout layer is trained on top, which avoids the instability of training quantum parameters directly.

The goal was explicitly complexity reduction rather than a constant-factor speedup on the existing method. The team reports the resulting algorithm is near linear in the number of bases, with the non-linear parameters small enough that the practical behavior sits very close to linear.

What the Benchmarks Show

The team benchmarked the approach against a deliberately broad set of comparisons, using a virus genome as the reference sequence:

Three scenarios were tested. Insertion takes a DNA segment, inserts adapter bases into it, and asks the method to align it back to the reference. Deletion removes bases and does the same. A genome benchmark takes portions of a genome and aligns them against the whole. Accuracy was scored on identity, which compares bases individually, and similarity, which groups bases and scores the alignment across those groups.

Across most of these tests, QRC landed at accuracy comparable to the classical baselines once error bars were taken into account, while achieving near-linear rather than quadratic scaling. On the mismatch scenario specifically, where individual bases are substituted rather than added or removed, QRC outperformed classical reservoir computing.

That combination, classical-comparable accuracy with a materially better complexity class, is the result worth paying attention to.

Where It Falls Short Today

Moraes was direct about the limitations, and they are worth stating plainly rather than burying.

Readout time is the current bottleneck. The measured wall-clock time does scale near linearly with the number of bases, which is the behavior the algorithm predicts, but extracting results from the quantum system adds enough overhead that the approach still loses on absolute time to the established classical aligners. The complexity advantage is real; the end-to-end timing advantage is not there yet.

Training the classical readout neural network introduces significant latency, but this step only needs to be performed once, after which the trained model can be saved for future use.

The tests were also run at small scale. The reservoir used eight qubits, a size constrained by what the team could simulate, and the deletion benchmark ran on segments of about 50 bases against a reference of a few thousand. Real crop genomes are many orders of magnitude larger.

The team's own reading is that quantum's relative advantage over classical reservoir computing should grow with reservoir size, since a classical reservoir has to be extended substantially to absorb longer inputs while a quantum reservoir represents them more compactly. At eight qubits the two sit close together, which is roughly what that reasoning would predict. But this is a hypothesis about where the curves diverge, not a demonstrated result, and the team is explicit that larger systems need to be tested before anyone can claim otherwise.

Why This Matters

For enterprise and research teams evaluating where quantum methods might eventually matter, this project is instructive for two reasons that have little to do with the hardware.

The first is that the leverage came from reformulation. Treating a genome as a temporal series rather than a static string is not a quantum insight, strictly speaking, but it is what opened the door to a quantum-native method and to a different complexity class. Teams looking for quantum-relevant problems often search for existing algorithms to accelerate. The more productive question is frequently whether the problem can be posed differently.

The second is that the honest accounting is what makes the result usable. A bottleneck identified precisely, at readout, is a bottleneck an engineering roadmap can target. A scaling hypothesis stated as a hypothesis is one that a larger reservoir can test. That is what early-stage applied quantum work is supposed to look like.

What Comes Next

The immediate work is scale: larger reservoirs, longer sequences, and benchmarks that approach the size of real crop genomes rather than viral references. Whether the projected divergence between quantum and classical reservoir computing materializes at those sizes is the question the next phase has to answer.

Alignment is also not the only place this framing could apply. Any domain with short, noisy, sequential records and a need to localize patterns within them shares the structural features that made this approach viable.

For QuEra, this is the kind of collaboration co-design is meant to produce: a partner with deep domain access, a real computational bottleneck with commercial stakes behind it, a testable hypothesis, and results reported with their limitations attached.