Molecular Quantum Solutions: Shaping the Future of High-Performance Cloud Software to Predict Chemical and Material Properties
As computing power increases exponentially, the biopharma and chemical industries are undergoing a transformative shift from trial-and-error to the purposeful design of experiments using computer simulations.
In the past, building and maintaining your own high-performance computing cluster was necessary to simulate drugs and materials accurately, a costly endeavor worth pursuing only for large chemical companies. With microprocessors becoming increasingly potent and the advent of scalable cloud solutions, smaller companies and labs can now access the necessary computing resources.
Molecular Quantum Solutions, commonly called MQS, leverages high-performance cloud infrastructure to democratize access to rapid and accurate thermodynamic chemical and material property predictions. In the summer of 2023, MQS raised a €700K pre-seed round from Scale Capital, Rockstart Emerging Tech, and Stokbro Invest.
Learn more about the future of high-performance cloud infrastructure to predict chemical and material properties from our interview with the founder and CEO/CTO, Mark Nicholas Jones:
Why Did You Start Molecular Quantum Solutions?
Throughout my undergraduate and Ph.D. studies in computer-aided chemical engineering at the Karlsruhe Institute of Technology (KIT) and the Technical University of Denmark (DTU), I wanted to automate things using software tools and programming languages. I never wanted to just work in the laboratory or with spreadsheets. I wanted to leverage Fortran, C/C++, and Python to develop sophisticated algorithms.
In computer-aided chemical engineering, much of the simulation, optimization, and modeling tools are closed-source (just think of Aspen HYSYS, AVEVA, gPROMS, GAMS, and COSMOtherm). One reason for keeping tools closed-source is that building computational models to predict chemical and material properties or to simulate manufacturing processes from fundamental physics principles is hard and costly. Developing and optimizing simulation packages in Fortran or C/C++ takes years to reach an industry-grade level.
Quantum chemistry has a stronger tradition of being open-source, and public code repositories on Sourceforge and GitHub have also pushed the more conservative chemical engineering industry to become more open. We can see this, for example, with the DWSIM process simulator, Pyomo, Jump, COSMO-SAC, and other packages.
Around 2015, more and more people started applying machine learning models in chemical engineering with the help of open-source and freely licensed packages such as RDKit, scikit-learn, or PyTorch. This made me realize it might be the right time to found a startup due to the advent of cloud infrastructure, quantum computers and the customer pain of deploying and applying open-source and freely licensed quantum chemistry packages.
MQS was founded at the end of 2019. We leverage such open-source packages and the latest machine learning models to make accessing thermodynamic prediction models easier.
How Do You Use Simulations to Predict the Properties of Molecules?
Numerical quantum chemistry predicts the properties of molecules and materials from first principles, i.e., the fundamental laws of quantum mechanics.
The challenge is that these quantum mechanical calculations are prohibitively expensive computationally. That’s why you need to make abstractions and simplifications that still capture the essential physics but allow predictions to be made more easily. You still need a lot of computing power, which wasn’t available ten years ago publicly in the cloud, and you had to build your own high-performance computing cluster if you wanted to run such simulations.
We develop new simulation algorithms in-house and build the software infrastructure to leverage existing open-source packages. The quantum chemistry simulation algorithms in open-source/free-license packages have been meticulously refined over the years, and no startup can invest the time and effort to rebuild all of those from scratch. That’s why we build software infrastructure to run them efficiently in the cloud. When you look through the code of the packages (e.g. PSI4, pySCF, MOPAC, Quantum Espresso), you can find interesting things, such as insightful comments from researchers who developed or implemented the algorithms back then. We sometimes feel like ‘archeologists of numerical quantum chemistry algorithms’.
We also incorporate advanced machine learning, like graph neural networks, that account for the graph structure of molecules to predict their properties. We use experimental and first-principles simulation data to train these models, and it is this combination of fast and efficient numerics with state-of-the-art machine learning where we found our unique edge.
How Do You Leverage High-Performance Cloud Infrastructure Efficiently?
One challenge is that quantum chemistry packages are often based on old legacy code but are still being used since it took a large amount of effort to implement them in the first place. Back when they were developed, fast prototyping languages like Python and large-scale parallel compute infrastructure were not yet readily available. Today, it is still not straightforward to parallelize these packages and make them run on different types of CPUs and GPUs or to interface the software tools with methods utilizing quantum processing units (QPUs).
We know other companies have put significant effort into compiling those packages for GPUs and demonstrate a certain speed-up. Relative to the implementation overhead and the costs of GPUs, we figured that it is not worth it for us to shift to GPUs for quantum chemistry packages already today. But we are preparing to leverage them in the future.
Instead, we compile quantum chemistry packages for Intel, AMD, or Arm-based CPUs, which keeps energy consumption and, thus, server costs low. This is quite unique; not many people know how to compile these packages for different processors and make them run efficiently. Distributing the computation across many CPUs works very well for efficient high-throughput screening of molecules.
We are continuously benchmarking our simulations on different chips and have also looked into FPGAs and ASICs. But making our simulations work on these more specialized chips requires even more work overhead than adopting GPUs. And finding FPGA engineers can be harder than finding quantum computing experts! As a startup, you need to consider talent availability—don’t develop exotic tech stacks where you can’t find people to develop and manage them.
Our goal is to design our software infrastructure well and make it backend-agnostic. This way, we can work with any backend that makes sense in the future, especially when it comes to various CPU, GPU, and QPU architectures. We realized that there is a real business opportunity to develop and run quantum chemistry simulations cost-efficiently and solve the customer pains involved in utilizing highly performant prediction pipelines.
At MQS, we have focused on our core competencies from the start and analyzed the algorithms and suitable architectures for each use case and calculation step in detail. As we have demonstrated in the past, we also take new developments into account when the timing is right.
How Did You Develop a Business Case for Your Simulations?
The value-add we envision with our prediction pipelines is to give our clients a well-informed analysis of their pharma or chemical formulation problem and which lab experiments are necessary to evaluate whether a certain mixture of components gives the desired overall property of the formulation.
Thanks to modern cloud infrastructure, we can provide our algorithms in an automated and containerized fashion on any cloud provider’s infrastructure or even on our customer’s cloud infrastructure. This allows us to offer our algorithms modularly and via a monthly subscription or usage-based. The latter has become feasible since cloud providers made usage tracking of containers/images possible, thus enabling flexible offerings based on usage metrics.
With all our tools combined, we have built a foundational framework for bridging the gap between quantum chemistry, thermodynamics, and chemical engineering. The MQS tool stack can address many different use cases, such as formulation development, helping green chemistry companies estimate the CO2 solubility in a specific solvent, or developing better catalysts to convert chemical compounds. We focus on niche applications (drug formulation development) with a beneficial price point to generate early revenue while preparing to go fully horizontal over time for multiple use cases within pharma, biotech, and chemical product discovery, formulation, and manufacturing simulations. This strategy de-risks MQS as a business while navigating toward true product-market fit.
How Did You Evaluate Your Startup Idea?
Being part of different accelerator programs, such as the Creative Destruction Lab Quantum stream, was immensely helpful in validating our business with experts in the field. The advisors helped us realize that to build an impactful company; we’ll eventually need to get into chemical synthesis.
Our current focus is building the software infrastructure to run quantum chemistry calculations through our cloud pipelines, where the main value-add for our clients is guiding them in designing the right experiments.
We have an EIC accelerator application pending that, if successful, will allow us to build our own lab—we have two master thesis students developing a blueprint of the lab—so we can perform experiments with robot arms to accelerate our customers’ R&D work. We collaborated with a large chemical company that faced an up to six-month bottleneck to conduct a single experimental study in-house, so we aim to do such experiments quicker and more efficiently than any company could do themselves.
Finally, we will become a synthesis company, synthesizing compounds that help our customers, e.g., deliver a drug more effectively to a specific region in the human body. Our end goal is to accelerate our customers’ R&D pipeline and product optimization from start to finish, e.g., validating a new drug delivery system initially through software simulations and then in the lab in a couple of days instead of months or years.
What Advice Would You Give Fellow Deep Tech Founders?
Realize your timeline as a startup, where you are with your idea, and if and when it is the right time to raise money from VCs. For example, developing quantum chemistry algorithms and software infrastructure to run them in an easy-to-use and tailored way takes years of development.
If you don’t see that your prototype is market-ready, be patient, take the time, stick to your current job, and develop your product on the side. When going full-time on your start-up project, start applying for equity-free soft funding and research grants to build a grant funnel and get your technology to a level where customers can and want to use it. If you choose to spin out / start a company with a prototyped pipeline as first-time founders, then be prepared for bootstrapping and ensure you have a skillful team.
Once you raise venture capital, the clock starts ticking. Launching a half-baked minimum viable product can be and will probably be a stressful and exciting endeavor. Be prepared to manage pivots, customer setbacks, kind/tough pressure from investors, continuous iterations, marketing, and business operations, both behind the scenes and “on stage.” You must become an all-rounder, developing many skills to get a company off the ground.
The venture capital world looks shiny, and many take the leap. However, the failure rate is high. In my opinion, deep tech founders have to lay a good, de-risked business foundation for scaling later on. Taking too many shortcuts, not focusing on product development, and/or focusing too much on flashy marketing can harm. Developing the product in close feedback loops with your customers, team, and trusted experts should be your highest priority. Ask yourself regularly:
- How sure are you that your idea is or will ever be market-ready, and how do you get there?
- Do you have the financial backing privately or through equity-free soft funding, such as grants, to develop your product while the market is developing? And are you testing market entry in different ways?
- Do you have the stamina to go through all the ups and downs?
- And how long are you willing to dedicate your life to this company, which will eventually cease to be ‘your’ company?
For me, MQS is more than a lifetime project; it’s a passion. This enthusiasm stems not only from MQS itself but also from my deep interest in the underlying theory and knowledge that spans software design, quantum chemistry, chemical engineering, computing technologies, machine learning, business infrastructure, business model development, strategic marketing, and quantum computing.
Want to Know More?
Learn more about how MQS leverages high-performance cloud infrastructure for quantum chemistry and materials science on their blog at https://blog.mqs.dk*
*Sponsored link – we greatly appreciate the support by MQS