Skip to content
Matmerize

Accelerating Polymer Formulations Design with Available
Raw Materials Using PolymRize™

A leading polymer formulations company initiated a proof-of-concept with Matmerize to explore the adoption of
AI-driven informatics for accelerating the design and optimization of resin formulations. The project showcased
how PolymRize™, a flagship product of Matmerize powered by AI and polymer informatics, can accelerate new
formulation design by leveraging extensive internal and supplier datasets of available raw materials. By
integrating ML-assisted predictive capabilities, the company was able to search the formulation space efficiently
and reduce development cycles for designing new high-performance formulations that meet target metrics.

Customer Challenge

A global leader in industrial coating resins and additives faced challenges in rapidly designing new resin
formulations that met specific target properties. The traditional formulation development approach requires
extensive experimental iterations involving effectively infinite combinations of available raw materials, making
the process time-consuming and resource intensive. The company sought an advanced data-driven approach to
efficiently screen and design new materials (based on available raw material ingredients) while optimizing their
performance.

Solution: PolymRize™ AI Platform

PolymRize™ enabled the company to adopt an AI-driven design approach to address key challenges faced by its
traditional design methodology. To validate this approach, three large internal datasets, each containing over
10,000 data points involving available raw material ingredients, were utilized to develop twelve data-driven
prediction models. The three datasets included the formulation names, quantities and combinations of various
raw materials (monomers, additives), and over a hundred processing conditions.
Matmerize employed a multi-step process as demonstrated in Figure 1, that included:
Data Curation & Preparation: Converting raw formulation data into machine-readable formats, making it
ready for informatics processing.
Chemical Fingerprinting: Applying our proprietary multi-length-scale chemical fingerprinting techniques
to numerically encode polymer structures and their chemical behaviors.
DATA CURATION FINGERPRINTING MODEL TRAINING DESIGN & SCREENING
Handcrafted
fingerprints
“Polymer
Genome”
Machine-crafted
fingerprints
“PolyBERT”,
“PolyGNN"
0.9 0.30.2 0.7 0.3 …
0.3 0.20.1 0.9 0 …
0.5 0.90.2 0 0.8 …
0.4 0.70.9 0.7 0.5 …
Figure 1. Data- and AI-driven design approach powered by PolymRize™, the flagship product of Matmerize
Predictive Model Training: Developing machine learning (ML) models to predict key properties, such as
glass transition temperature, viscosity, color, pH scale, and epoxy equivalent weight. A total of 12 target
properties and performance ranges were identified.
Design & Screening: Leveraging ML models to generate and screen thousands of new polymer
formulations, identifying novel formulations that best met the desired performance criteria.

Key Benefits

The Matmerize approach enabled a more data-
driven and efficient design process, reducing both
time and cost for optimizing new resin formulations
to meet performance targets. It also significantly
enhanced the company's innovation capacity by
accelerating the exploration of new combinations to
achieve targeted performance criteria.
As shown in Figure 2, the project effectively
demonstrated the value of AI-driven design for the
company through four key benefits: enhancing
predictive accuracy with reliable ML-based property
predictions, optimizing formulation design via
efficient candidate screening, implementing active
learning cycles to continuously refine models with
new data, and developing custom AI models
embedded with domain expertise and constraints.

Key Takeaways

Targeted Formulations Design: A key takeaway was the AI models' accuracy, which facilitated the rapid
design of optimal resin formulations through the virtual generation and evaluation of thousands of candidates
in mere minutes. The models empowered users to explore a vast space of monomers, additives, catalysts, and
processing conditions—driving faster, more efficient, and highly targeted formulation innovation.
Scalability & Future Implementation: The project demonstred that AI-driven formulation design and
development can be scaled effectively, leveraging the extensive internal raw datasets the company had
accumulated over the years.
Data Infrastructure Development: Robust workflows were established for ongoing data collection and
curation, as well as design of experiments (DOE), allowing for the automated updating of predictive models,
ensuring continuous improvement.
User-Friendly Integration: R&D teams were able to easily adopt PolymRize™ via a browser-based interface
and APIs integration into their existing workflows, ensuring smooth adoption.

Conclusion

The project highlighted the transformative power of polymer informatics in formulation design. By integrating
PolymRize™ into its R&D workflow, the company is now positioned to scale innovation in new formulations,
significantly reducing development time while meeting or exceeding key target property requirements.
Trained ML models provided highly reliable
property predictions, reducing the need for
extensive lab testing
ENHANCING
PREDICTIVE
ACCURACY
AI-generated candidates streamlined the multi-
objectives screening process, identifying
promising resin formulations faster
OPTIMIZING
FORMULATION
DESIGN
AI models iteratively improved with new
experimental data, refining predictions and
expanding the chemical design space
IMPLEMENTING
ACTIVE LEARNING
CYCLES
The project established a foundation for
embedding specific constraints and domain
expertise directly into the PolymRize™,
preserving valuable institutional knowledge
CUSTOM FEATURES
FOR THE
COMPANY
Figure 2. Key benefits of AI-driven approach with PolymRize™