Researchers from Quantinuum and Amgen have conducted a study on using quantum computers for computational biology. They built quantum models to perform sequence classification relevant to the design of therapeutic proteins. The models were run on emulators of state-of-the-art noisy quantum processors and then on the Quantinuum H1-1 trapped-ion quantum processor. The results were in close agreement with noiseless exact simulation. This work is the first application of near-term quantum computing to a task critical to the design of therapeutic proteins, suggesting potential for larger-scale applications in this and related fields. The work is published in Arxiv, “Peptide Binding Classification on Quantum Computers”.
Quantum Computing in Computational Biology
Quantum computing is being explored for its potential applications in computational biology. In a recent study, researchers used quantum models based on parameterised quantum circuits to perform sequence classification, a task relevant to the design of therapeutic proteins. The performance of these quantum models was found to be competitive with classical baselines of similar scale.
The study also examined the effect of noise on the quantum models by running them on emulators of state-of-the-art noisy quantum processors. Error mitigation methods were applied to improve the signal. The quantum models were then executed on a trapped-ion quantum processor, and the results were found to be in close agreement with noiseless exact simulation.
The researchers also performed feature attribution methods and found that the quantum models identified sensible relationships, at least as well as the classical baselines. This work is the first proof-of-concept application of near-term quantum computing to a task critical to the design of therapeutic proteins.
Quantum Machine Learning (QML)
In the rapidly growing field of quantum machine learning (QML), parameterised quantum circuits (PQCs) are being used as machine learning models. PQCs offer several advantages, including relatively easy implementation on quantum hardware, sufficient expressive power for various tasks, and the ability to be trained with a classical machine learning objective function and optimisation loop.
In this study, PQCs were built and trained to solve a problem in the domain of computational biology. The task was binary classification of peptides, which are short chains of amino acids, according to their binding affinity to a target molecule. Peptide binding plays a crucial role in cellular signalling, protein trafficking, immune response, and oncology, and predicting their binding affinity is a long-standing challenge.
Quantum Models and Classical Baselines
The performance of the quantum models was compared against that of classical baselines, and the quantum models were found to perform as well as the classical neural models. This is a positive outcome, given that quantum models are still in their infancy and face significant technical challenges such as noise and error rates.
The inherently different nature of quantum computing from classical computing implies that the strengths of the two paradigms might lie in different problem domains. Therefore, demonstrating comparable performance on a typical machine learning task offers encouraging insights into the potential of quantum computing in practical applications.
Quantum Models on Quantum Processors
The modest size of the quantum models allowed the researchers to execute them on quantum processors and correct the effect of noise using standardised error mitigation methods. This established the first proof-of-concept experiment involving the application of quantum models to a simple computational biology task on currently available quantum hardware.
The researchers analysed each individual amino acid’s contribution to the binding probability, and observed that the small-scale PQC-based models recovered this information at least as well as the classical baselines of similar scales.
Quantum Models in Computational Biology
The study conducted an extensive investigation into using quantum ML models on a computational biology task. A methodology was detailed that allows the representation of sequence models on quantum hardware. The results from a proof-of-concept experiment on quantum hardware showed the potential of quantum models in the field, by achieving results similar to classical baselines.
The task chosen for the study was binary classification of peptides according to their binding affinity to a target molecule. This task is critical to the design of therapeutic proteins, and the successful application of quantum models to this task opens the route toward larger-scale applications in this and related fields.
“In the rapidly growing field of quantum machine learning (QML), the use of parameterised quantum circuits (PQCs) as machine learning models (Benedetti et al., 2019) has found a wide range of applications. PQCs provide a number of advantages as tools for QML, the most important of which are relatively easy implementation on quantum hardware, sufficient expressive power to be applied in several tasks (Du et al., 2020), and perhaps, above all, the ability to be trained with a classical machine learning objective function and optimisation loop.”
Charles London, Douglas Brown, Wenduan Xu, Sezen Vatansever, Christopher James Langmead, Dimitri Kartsaklis, Stephen Clark, Konstantinos Meichanetzidis
“Peptide binding plays a crucial role in cellular signalling, protein trafficking, immune response, and oncology, and predicting their binding affinity is a long-standing challenge (Das et al., 2013; Wang et al., 2019). Due to the importance of the peptide-MHC interaction for adaptive immunity and the large datasets available for training, in the current study we focus on this peptide binding problem.”
Charles London, Douglas Brown, Wenduan Xu, Sezen Vatansever, Christopher James Langmead, Dimitri Kartsaklis, Stephen Clark, Konstantinos Meichanetzidis
Summary
Researchers have successfully used near-term quantum computers to perform sequence classification in computational biology, specifically in the design of therapeutic proteins. This constitutes the first proof-of-concept application of quantum computing to a task critical to the design of therapeutic proteins, demonstrating comparable performance to classical computing and opening the route toward larger-scale applications in this and related fields.
- A team of researchers from Quantinuum and Amgen have conducted a study on using near-term quantum computers for tasks in computational biology.
- The team constructed quantum models based on parameterised quantum circuits to perform sequence classification, a task relevant to the design of therapeutic proteins.
- The quantum models were run on emulators of state-of-the-art noisy quantum processors, and error mitigation methods were applied to improve the signal.
- The models were also executed on the Quantinuum H1-1 trapped-ion quantum processor, and the results were found to be in close agreement with noiseless exact simulation.
- The study constitutes the first proof-of-concept application of near-term quantum computing to a task critical to designing therapeutic proteins.
- The researchers believe this opens the route toward larger-scale applications in this and related fields, in line with the hardware development roadmaps of near-term quantum technologies.
- The study also found that the quantum models identified sensible relationships, at least as well as the classical baselines.
- The researchers concluded that the results offer encouraging insights into the potential of quantum computing in practical applications.