How to use Quantum Boost to accelerate the development of chemicals and materials

What is the Quantum Boost Platform

The Quantum Boost platform gives formulators the tools to improve their experimental strategy, from setting project objectives to figuring out the best experiments to conduct. With advanced analytics and suggestions from AI, the platform helps you move quickly and accurately toward your target formulations. This tutorial will show you the main features of Quantum Boost and how to get the most out of the platform to speed up your projects.

How does it work?

Quantum Boost uses advanced Artificial Intelligence algorithms, particularly Bayesian Optimization, to navigate your complex factor space efficiently. Unlike the usual Design of Experiments (DoE) software, which can fail when dealing with problems that are high-dimensional, non-linear, or non-convex, Quantum Boost works well even in tough situations. By integrating your existing experimental data into its optimization model, the platform helps you reach your project objectives swiftly and accurately. For a closer look at how Bayesian Optimization compares to DoE, you can read our Bayesian optimization vs. DoE article.

How to Define a Project in Quantum Boost

When defining a project on Quantum Boost, you are setting up a roadmap for your experimentation. It is important to provide the system with as much accurate information as possible to leverage its AI capabilities fully. Here is a step-by-step guide on how to define a project:

1. Set Your Goals

Goals in Quantum Boost are the outcomes you are aiming to achieve. These could be attributes like "Viscosity," "Surface Tension," or "Resistivity."

• Response: What you are trying to affect (e.g., Viscosity)
• Goal: What you want to do with the response (Minimize, Maximize, or Match target)
• Target Value: The specific numerical aim you are shooting for (e.g., 2Ω.m for Resistivity)

You can add as many goals as you want, and if you manage to reach your targets, Quantum Boost will continue to optimize to make those goals even better.

2. Define the Factors

Factors are the variables that you can control or monitor during your experiments. There are three types:

• Ingredients: These are the components that make up 100% of your formulation.
• Continuous Variables: These are process parameters like temperature, pressure, or flow rate.
• Categorical Variables: These are discrete options like mixing method or type of catalyst.

Factors have some considerations based on each type:

• Controlled: Whether the factor is within your control. While this doesn’t apply to ingredients (they are always controlled), environmental factors like humidity might not be. When you analyse the data in our analytics, adding uncontrolled factors can help you understand their impact on your experiment outcomes.
• Constant: If the factor remains unchanging throughout the experiments. This is useful when a certain component must always make up a specific percentage of the formulation or if a process parameter is fixed.
• Range for Suggestions: This is important for 'Ingredients' and 'Continuous Variables.' You can specify either a constant value (discussed above) or a range. For example, an ingredient could range from 0% to 70% in the formulation, or a continuous variable like temperature could range from 20°C to 60°C. This range sets the boundaries for the algorithmic suggestions, ensuring they align with practical constraints.

The beauty of Quantum Boost lies in its flexibility. All these factors can be modified during the lifespan of your project without losing valuable data.

3. Adaptive Goals and Factors

One standout feature of Quantum Boost is its adaptability. You can change goals and factors mid-project without compromising the integrity of the existing data. This is incredibly useful in real-world scenarios where supply chain issues might force you to swap out ingredients or adjust your goals.

(Note: when dropping an ingredient used in past experiments, you would just set its value to a constant zero.)

By thoughtfully defining your project with accurate goals and factors, you are laying a strong foundation for successful experimentation. Remember, the more precise your initial setup, the more effectively Quantum Boost will steer you toward your objectives.

4. Implement Constraints: Sum and Group

Constraints are rules or limitations you set to guide the algorithm's suggestions further. They ensure that the experiments align with real-world conditions or specific goals you are trying to achieve. Quantum Boost allows you to set two types of constraints: Sum Constraints and Group Constraints.

Sum Constraints:

These constraints deal with the collective effect or total sum of a particular group of factors in your formulation. For example, you might want to ensure that:

• The total amount of solvents used in your formulation should be less than or equal to 40% of the final solution.
• The total cost of your formulation must not exceed $150 per 100g.

These constraints ensure your formulation meets the practical limitations and pre-defined project specifications. You can define as many sum constraints as necessary to narrow your experimentation to the most useful and feasible options.

Group Constraints:

While sum constraints focus on the collective quantities or costs, group constraints limit the selection from a particular category of ingredients. Examples include:

• Opting for only one of the three available solvents.
• Limiting the usage to a maximum of two additives.

Group constraints are particularly useful when you have mutually exclusive options or want to limit complexity by specifying the maximum number of components used from a specific category.

Together, these constraint functionalities in Quantum Boost empower you to conduct scientifically rigorous, realistic experiments aligned with your business or research objectives.

Adding past experiments

The core of Quantum Boost's algorithm relies on your past experiments. By analyzing these experiments, we build a model of your factor space to suggest the most impactful next steps toward your objectives. Here is how to add experiments to fuel this model:

Method 1: Adding New Experiments

You can manually input new experiments from scratch, filling in details about factors and responses.

Method 2: Using Existing Experiments

If your organization has completed relevant experiments in other projects, you can import these directly into your current project.

Filling in Details

After experiments are added, input your factors and their corresponding responses. The platform provides visual cues:

• Response cells will turn green when goals are met.
• Sum constraints columns will turn red if your formulation violates any set constraints.

Important Considerations:

• Missing ingredients are implicitly considered as 0% and do not require explicit input.
• The current algorithm cannot interpret missing values for continuous or categorical factors, and these experiments will be ignored. This is important because, for instance, a missing temperature value can't be assumed to be zero degrees.

By carefully inputting your experiment details, you contribute to the precision and effectiveness of Quantum Boost's model, streamlining your journey toward meeting your project objectives.

Generating suggestions

Generating suggestions is at the core of what makes Quantum Boost so efficient. The platform uses its advanced algorithms to pinpoint the most promising next experiments for you to perform based on your project definitions and previous experiments. The aim is to lead you to your desired formulation with the fewest experiments possible.

Here are some key aspects to consider when generating suggestions:

• Exploration vs Exploitation: Quantum Boost intelligently balances between exploring new, uncharted areas in your factor space (exploration) and optimizing based on the data it has already gathered (exploitation). This balance ensures you get the best of both worlds: new discoveries and refined results.
• Setting Quantity and Unique Suggestions: You can specify the number of suggestions you'd like to receive, with limits varying based on your subscription tier (up to 10 for enterprise/free trial and 2 for the starter tier). While you can generate suggestions multiple times, you will encounter the same ones repeatedly without changing the project. A greater quantity of suggestions ensures you can generate a larger unique set, enabling faster parallel exploration of the factor space.
• Optimal quantity: If your project comes with no past experiments, the most optimal quantity to choose is the square root of the number of non-constant factors in your project. E.g., if you have 16 non-constant factors, you should start by generating 4 suggestions. This is the minimum required for our algorithms to learn basic patterns from the data. Once there are past experiments in the system, it is best to have as tight of an iterative loop as possible. Ideally, you would generate a single suggestion, perform the experiment, input back its results, and generate the next suggestion. This ensures that each single experiment has access to more information than the last. Of course, if your lab setup allows for parallel experimentation at minimal cost, you should generate more than one suggestion at a time, but still try to keep the iteration loop as tight as possible.

By keeping these considerations in mind when generating suggestions, you are positioning your project for both accelerated progress and optimized resource utilization.

Analytics: Insights Made Easy

The analytics section in Quantum Boost is designed to provide user-friendly insights tailored for formulation scientists. Unlike traditional DoE software, which often requires a statistical background to interpret, our analytics are straightforward and intuitive. Below are the key types of analytics provided:

Scatter Plot

This graph offers a visual way to identify correlations among factors, responses, and constraints. Key indicators on the graph include:

• Pareto front: A dotted line highlighting optimal trade-offs between your chosen minimize and/or maximize responses.
• Target lines: Red dashed lines represent any set targets or bounds for your factors or responses.

Factor Effects Plot

This plot informs you about how various factors impact your responses. A factor that doesn't appear for a given response is likely not statistically significant. Reading the information popup associated with this graph is advised for a more nuanced understanding.

Model Reasoning Plot

The Model Reasoning Plot is designed to help you better understand the underlying logic that guides the algorithm's suggestions. The graph represents what the algorithm thinks about the relationship between the selected factor and response, complete with a band of uncertainty.

For an in-depth understanding of each plot, click on the information icon associated with each graph. This will provide detailed explanations and use-case scenarios to help you maximise these analytical tools.

Conclusion

Quantum Boost presents a robust platform for any formulator aiming to expedite the process of material and chemical development. Understanding and utilizing its multifaceted features allows you to stay ahead of the curve in your research and development endeavours.

If you have not already, sign up for a 14-day free trial today!

To learn more and go through a video demo of the platform, click here!

‍