Quantum Genomics

Predictive text,for DNA.

The same logic behind your keyboard's next-word suggestion — applied to genomic sequence. This study evaluates two quantum approaches for predicting DNA sequence against classical Markov and Transformer baselines.

Dat Chi (Ryan) Le · Abubakar (Sid) Iliyasu · Gyanateet Dutta

Presented for the International Year of Quantum Science and Technology, with the University of Bradford, Quantinuum, and Bradford 2025 UK City of Culture.

01 — The Idea

From Markov chains to transformers, now to qubits.

Sequence prediction has moved from simple Markov-style transition models to transformer architectures like BERT and GPT. Genomic sequence prediction has real medical stakes — identifying causative mutations, therapeutic targets in tumours, drug response, and disease risk — which makes the case for exploring a quantum approach worth testing directly against those classical methods.

02 — Data & Encoding

256 tokens fit exactly into 8 qubits.

DNA sequences from NCBI are broken into 4-letter tokens (4-mers) using a sliding window, giving a vocabulary of 256 unique tokens — enough to capture local motifs like codons or regulatory sites, and small enough to map onto 8 qubits without wasting hardware.

4-mer

Tokenisation of raw DNA sequence

256

Unique tokens (4 bases ^ 4 positions)

8

Qubits — log₂(256), an exact fit

NCBI

Source dataset for genomic sequences

03 — Two Quantum Approaches

A sampler, and a quantum transformer.

Sampler

QMCMC (QD-HMC)

The same quantum Hamiltonian Monte Carlo sampler used in our financial forecasting work, applied here to Bayesian regression over tokenised genomic sequence — generating calibrated probabilities for the next token via quantum coherent evolution and Metropolis–Hastings acceptance.

Quantum Transformer

Quixer (QNLP)

A quantum analogue to attention, built from a Linear Combination of Unitaries and the Quantum Singular Value Transform (LCU + QSVT). Token embeddings drive a parametrised circuit whose quantum state scales as O(N) with sequence length, against O(N²) for classical attention.

04 — Results

~90% accuracy, ahead of classical baselines.

Both quantum approaches outperformed the classical Markov chain and Transformer baselines, which reached roughly 80%. Quixer with a hybrid classical-prior initialisation (“knowledge injection”) also converged faster than a standard quantum initialisation.

QMCMC (quantum)
90.6%
Quixer (quantum)
90%
Classical baseline
80%

QMCMC also scored 67.0% token accuracy and 0.669 F1; Quixer's figure is approximate, read from validation-accuracy curves.

05 — Impact

Mapped against the UN Sustainable Development Goals.

3

Good Health & Well-Being

Earlier diagnosis and personalised medicine through better sequence prediction.

9

Industry, Innovation & Infrastructure

Quantum-classical innovation for next-generation genomic modelling.

17

Partnerships for the Goals

Interdisciplinary collaboration across universities and industry.

06 — Conclusions & Next Steps

Feasible today; next, real hardware and longer sequences.

Next steps: running on actual quantum hardware (Quantinuum's H-Series), extending to longer 6-mer and 8-mer tokens, chromosome-level sequence, and applications in drug-target interaction, cancer genomics, and metagenomics.