Key Takeaways

• Molecular dynamics simulations and descriptor based machine learning bring deeper chemical insight into R&D work.
• Recent advances in agentic AI now allow specific modeling workflows to be carried out using natural language prompts combined with literature based datasets, reducing reliance on experimental input.
• Raw materials suppliers to the coatings industry are already embracing advanced modeling and simulation.

As a polymer chemist by training, I never felt any real need for computational chemistry. In every role I’ve had throughout my career, it simply wasn’t part of the toolbox. To me, computational chemistry belonged in academia; something for theorists, not for people running experiments in a lab. I assumed hands on work would always be more useful. I didn’t see the need for it, and I certainly didn’t expect to use it.

Then again, I once felt the same way about AI.

So when should we start incorporating computational chemistry into our work?

Now. It’s time.

I recently attended a two-day conference hosted by Schrödinger titled “Frontiers in Digital Chemistry.” Schrödinger, founded in 1990, is a software company offering a physicsbased computational platform for predicting molecular properties across life sciences and materials science. While they’re best known for their life science solutions, their materials science segment is growing, and that was the focus of the conference.

Since the event was nearby in New York City, I decided to take a chance and attend. I figured if the content went completely over my head, I could always slip out early. Instead, I found myself fully engaged for two days of talks from both Schrödinger scientists and customers, capped by a demo of their new AI agent, Bunsen, which can execute complex workflows from naturallanguage prompts.

Several presentations used molecular dynamics simulations to show how molecules interact. For example, one presenter touched on how plasticizers interact with polymers. Seeing these interactions visualized was eye opening. For a synthetic or formulation chemist, watching a simulation of two molecules engaging is far more impactful than reading a paragraph about it. It leaves a lasting impression and immediately suggests next steps. In coatings, where we constantly care about how binders interact with other formulation components, these simulations felt directly relevant. I found myself thinking of multiple situations where this capability would be incredibly useful.

Other talks focused on descriptor‑based machine learning models used to identify replacements for polymers with very specific haptic properties. These models were trained on descriptors, which are numerical values that capture chemically meaningful features such as solubility parameters, molecular weights, etc. Building these models requires real chemical intuition; understanding the underlying chemistry and selecting descriptors that matter. Once trained, the model relates the target property to the descriptors or chemical properties, effectively embedding chemistry into the prediction engine. 

The summit concluded with a live demo of the Bunsen AI agent. Attendees were encouraged to interact with it using natural language. One participant asked Bunsen to build a machine learning model for melting points of a class of organometallic compounds. Within minutes, Bunsen had searched the literature, identified roughly twenty compounds with reliable melting point data, and constructed a model, all without a single experiment. It was a powerful reminder of how chemical knowledge, paired with what already exists in the literature, can dramatically accelerate lab work.

The summit drew strong attendance, including several raw material suppliers serving the coatings industry. And if these suppliers are already embracing advanced modeling and simulation, it raises an obvious question for the rest of us in the coatings space:

Should we be using these tools too?

It’s time.

#moleculardynamics #machinelearning #agenticai #schrodinger #digitalchemistry #innovation #coatings #chemicals #consulting

(photo credit: image generated by Microsoft Copilot)