Quantum technologies have long been pitched as a way to fundamentally change the way drugs are discovered; to start putting the theory to the test, researchers from pharmaceutical company GlaxoSmithKline (GSK) have been toying with top-notch quantum devices, comparing the methods put forward by IBM and D-Wave to get a better picture of what to expect from those leading the quantum race.
The conclusion? The method used by D-Wave, called quantum annealing, can already compete against classical computers and start addressing realistic problems; on the other hand, gate-based quantum computers, such as the one that IBM is building, remain short of enough qubits to run problems that are relevant to the real world.
All is not lost for gate-based methods – quite the contrary, in fact. GSK’s researchers foresee that the expected increase in qubit count in computers like these will allow quantum devices to show a significant performance advantage over classical hardware, for pharmaceutically-relevant life science problems, but also many other types of application.
The results of the scientists experiments are still in pre-print, and are yet to be certified by peer review; in addition, the trials only focus on a specific problem – the use of quantum computing to assist drug discovery. Nevertheless, the research offers a valuable overview of the capabilities of quantum devices as they stand, and of the limitations of different approaches to quantum computing.
The problem addressed by the scientists is well-established in classical computing. Called codon optimization, it consists of finding sequences of genetic code, called codons, that will ultimately lead to the expression of a particular gene. Up to six codons can be required to represent an amino acid, which in turn form the proteins that determine the gene.
In classical computing, codon optimization is addressed with genetic algorithms (GAs) that sample and iterate many different combinations of codons before settling on the most “optimal” solution. Due to the limited capabilities of the hardware, however, GAs cannot sample a large number of solutions in little time, which is why drug discovery is a lengthy process.
“Thorough sampling of the solutions space is therefore often intractable with biologically relevant use-cases,” wrote GSK’s researchers.
Quantum computing, however, and the ability of qubits to carry out various calculations in parallel, shows a lot of promise for this type of optimization problem, and would allow for a larger solutions space to be explored much faster.
This is why the researchers set out to investigate the potential impact of quantum computing for codon optimization. Using a quantum algorithm, called the Binary Quadratic Model (BQM) that can run on different quantum platforms, the team decided to test two markedly different models: D-Wave’s quantum annealing method, and IBM’s gate-based quantum computer.
D-Wave’s technology, found the researchers, holds a lot of potential. The Canadian company’s 5,000-qubit Advantage system was used to run the BQM; the system was capable of mapping 30 amino acids, and when compared to the classical algorithms, it was found to achieve similar results. “(The computer) is found to be competitive in identifying optimal solutions, and future generations (…) may be able to outperform classical GAs,” concluded the scientists.
Current generations of quantum hardware are not mature enough to surpass classical computing for problems such as codon optimization. In other words, D-Wave’s processor did not run the calculation better than a classical algorithm; but it proved that a quantum device could perform competitively, even on a life-size problem. As the technology increases in scale, the researchers expect it to eventually outperform classical techniques.
In separate experiments, a similar conclusion was reached by researchers at materials design company OTI Lumionics, which is banking on quantum technologies to develop electronics with new properties. Using an optimization algorithm that is similar to the one run by GSK’s scientists, OTI Lumionics designed a new electronic material that will let phone and laptop manufacturers build transparent, bezel-free OLED displays.
Just like GSK’s team, OTI Lumionics’ researchers looked at the performance of different quantum approaches when running the algorithm. They eventually settled on D-Wave, finding that, contrary to other cloud-based quantum services, the company’s processor could already compete against classical methods, and reach a degree of industrial relevance.
D-Wave’s quantum annealing processor, however, is only reflective of one particular branch of quantum computing: based on a system that is capable of optimizing itself to reach the lowest energy state, quantum annealing is only suited to specific optimization problems. On the other hand, it is much easier to operate and control than gate-model computers like IBM’s. For this reason, D-Wave’s quantum computer already boasts thousands of qubits, while IBM has only hit 65 qubits.
To compare the two methods, GSK’s scientists ran the optimization algorithm on IBM’s 24-qubit Qasm simulator. This limited the outcome to four amino acids; in addition, the performance of the device was variable, with many examples of the quantum algorithm returning invalid results.
According to the paper, modelling biologically-relevant sequences would require thousands of qubits with high connectivity. “Implementing a version of this program for IBM Q devices, while successful, shows that modelling practical systems requires too many qubits to be run on even the most advanced gate-based devices available (e.g. IBM’s 65-qubit Hummingbird device),” wrote the researchers.
But although they are currently less mature than quantum annealers, gate-based quantum computers are expected to significantly increase their qubit count, while also reducing error rates. IBM’s roadmap for scaling quantum technology, for example, anticipates that a 1,000-qubit system will be available by 2023.
“While current generations of devices lack the hardware requirements, in terms of both qubit count and connectivity, to solve realistic problems, future generation devices may be highly efficient,” said the researchers.
When this moment comes, the gate-based devices may be able to solve large-scale optimization problems – but also in running different types of calculations, from financial modelling to weather forecasting through traffic optimization. The range of applications that gate-based quantum computers will find, in fact, is likely to exceed that of quantum annealers. D-Wave and IBM told they didn’t want to comment on the research.
So, while D-Wave’s quantum processor is already making strides in solving real-world problems now, a new comparison will only be fair once devices like IBM’s catch up on hardware scaling; the strengths and weaknesses of different methods will be clearer then. Until then, you can expect plenty more compare-and-contrasting from curious scientists trying to get a peek of the future.