Accessible Computational Chemistry

For polymer chemists and formulators

If you’ve ever done synthetic polymer product development specifically for customers or potential customers then you’ve experienced what I think of as “the speciality chemical development process.” It can be akin to a hellscape if managed poorly. I’ll try to illustrate it for you in a short hypothetical scenario.

Let’s say a big customer comes to us and they are currently buying a product from our competitor. This is a customer that your company has wanted to land for years and this is the first time they have reached out asking for a meeting. Everyone is excited. The CEO shows up, all the R&D leadership are there with the senior marketing people, and your company’s general counsel is there too. It’s basically a scene from Succession.

As a newly minted PhD holder in polymer chemistry you are kind of stuck sitting at your desk wondering when everyone is done with their fancy lunch so you can swoop in and get those leftovers (grad school habits die hard). The meeting ends and the executive team drives off with the executives from the other company in some limos to a private airfield and your manager comes over to you as you are eating a leftover turkey sandwich and lays out the scenario to you:

This customer wants a custom designed polymer that doesn’t exist. They want better performance and want it to be better than what they are buying. They are willing to pay higher costs for better performance. Sustainability is a plus, but not necessary. Ideally, your management would like you to get these potential customers samples from the lab to try in a few months. The internal cost to produce shouldn’t be crazy and it should be compatible with the existing manufacturing capacity. And it needs to be compatible with the customer’s manufacturing process. And it should be patentable too, so make sure you don’t infringe on anyone else’s patents. And, if we need to do a PMN to the EPA let’s get that figured out as soon as possible since it takes forever to get something approved. Oh, can you go to Brownsville, Texas in two weeks to take a tour of their plant to get a feel for the process? Also, I need you to help support this other customer on this other project too because Murthy is leaving next week so get up to speed before he’s gone.

In the end, it’s on you as the polymer chemist to take all this in and figure out a plan on how to accomplish these needs from both your customer and your company. Where should you start synthetically and how many resources do you have at your disposal? It’s basically like doing a whole PhD all over again, but with a little bit of a mini-MBA and mini-patent law course thrown in there as well.

If it’s a serious project worth millions of dollars you might get to work on it full time with a full time technician or maybe a part time technician. If this feels like an under-resourced project to you then trust your instincts, but you need to understand a bit of the overall thinking that your executive team has considered. The problem with these “pie in the sky” projects, potentially worthmillions of dollars, maybe tens of millions, per year in additional revenue is that they tend to take 5+ years to close if they close at all.

The question being pondered: Is it worth working on for 5 years and what if it never finishes? If you cost $100k/year in salary and your technician costs $50k/year and you two work on it full time for five years that’s at minimum $750,000 in just R&D salary, not to mention lab space overhead, and the other people needing to support you in the effort such as engineering, legal, sales, marketing, and procurement. Let’s round up and say about $1.5 to $2.0 million over five years of costs with a high potential of no chance to win revenue.

Hypothetical Plan of Attack

You can get quite a bit done with a really small team of well trained chemists and engineers, but it takes longer than anyone wants. You might run 3 different hypotheses in parallel in the lab experimentally and creates a lot of work upfront, but I think as the technical lead you are trying to see what might work and what might not work. Having 1 out of 3 initial hypotheses work is a huge success and having all three fail is also somewhat a success. At least you know what not to do.

In addition to this stuff you’ve got people (not in the lab) asking, “when are we going to make progress on this and when can we realize sales?” Hopefully those people also have not promised your executive team a specific amount of money by a certain date without knowing if/how/when you will be successful. This can set a project up for failure, I would know because it has happened to me. If you spend 6 months on it and everything you’ve tried isn’t even coming close to working you also should not be afraid to advocate that it might not be possible and to kill your own project. This might feel dangerous to your own career if this is all you are working on 100% of the time. Also, make sure you aren’t 100% on some pie in the sky growth project.

The fastest I’ve seen a custom development process take is about 2 years. This isn’t so much about the skill of the researcher, the willingness of the customers, or market conditions, but rather you get lucky in the lab and the first few things you try tend to work. Luck can also be somewhat replicated through a mixture of good hypothesis driven research, a design of experiments approach, and a researcher understanding product cost structure. An ability to know when a specific hypothesis won’t pan out and then trying something new with new information is just as important as being able to come up with a good hypothesis in the first place. Also, knowing how unfeasible 90% of the potential routes you could take due to raw material costs, timeline constraints, CAPEX constraints, or regulatory constraints is also part of focusing your efforts on things that might yield viable results. 

But what if you aren’t lucky or supremely skilled and you are just like a regular research scientist trying to get by (aka, you don’t read newsletters like this one)? 

Technology… Maybe?

A lot of management consultants talk and write about digitization of the chemical industry from how data is generated and captured from in the lab all the way to the plant and your customers. There are laboratory information management systems like Dotmatics and customer relationship management software like Knowde. When it comes to simulation software for the chemical industry and R&D though I think most people are still just using ChemDraw, which is good for predicting ¹H NMR spectra, calculating molecular weight, but not great for telling you if a specific monomer will have an effect on the glass transition temperature that you want.

Spending a few weeks thinking and looking at patents and academic literature is critical for not wasting time in the lab. But what if you could model out a few avenues of research?

Molydyn

I got a chance to talk to Matthew Bone back in February about his software startup Molydyn. Matt and his start-up are trying to make molecular modeling of polymeric systems broadly accessible to polymer chemists. A recent perspective paper in Macromolecules summarized it best:

With recent advances in computing power, polymer simulations can synergistically inform, guide, and complement in vitro macromolecular materials design and discovery efforts.

Essentially the way Molydyn works is you can draw out your systems in a ChemDraw like interface, describe how many additional molecules you’d like to be simulated out from that starting system, some model parameters like initial density, and pick your forcefield type which defines the physics.

Matt’s view is that modeling polymeric systems just shouldn’t be that complicated. Matt described Molydyn’s platform Atlas to me as an accessible tool for molecular dynamics, a modeling technique that is less accurate than density functional theory (DFT) but can model large polymer networks and can be used for QSPR. Matt’s goal is to get people from drawing structures and spending months in the lab figuring out stuff the hard way to modeling systems within 10-15 minutes. It was pretty easy to generate the data when I tried it recently. Making sense of that data requires modeling it in a LAMMPS model. You can check out the Molydyn tutorial here:

Matt’s vision for the future of Molydyn is a fully realized experimental modeling system where users can draw, experiment, and get real time data in their hands. In the next update you won’t have to use an external modeling software and it will all be contained within their software. Molydyn will allow experimentalists to draw and model how monomer structures influence polymer properties and better understand structure-property relationships. From making rubbers to acrylics to epoxy resins and even formulations.

I think most companies would be willing to pay for something like Atlas, especially if it’s more cost efficient than a slightly better version that broadly exists right now. I think there is a bigger market though for something powered by machine learning or artificial intelligence.

The End Game

In my discussion with Matt I told him that I thought if he could develop a product that would allow chemists to train a machine learning algorithm he might really be onto something. In my experience chemical companies are typically specialists in a few things and that’s all they care about. A software program that can enable chemists to model whatever their heart desires is nice, but doesn’t allow them to leverage experimental data with simulation.

As a former practicing polymer chemist I would have loved to model out some systems, actually do the experiment, and feed my observed data back into the model, and then have that model get a little bit better. Maybe do some more modeling and go down the proverbial rabbit hole of trying to optimize a polymer system for a specific set of properties. Sometimes, I think back to the work I did and how it’s likely just collecting dust in a lab notebook. My failures and successes are forgotten (unless they were commercialized), but I’d like to think if you could use that data and train an artificial intelligence or machine learning algorithm maybe it could live on and still provide a bit of value.

Matt definitely sees this as the future, but first just getting modeling tools into the hands of experimentalists is their first goal. I’m sure there are warehouses full of old data or notebooks that could be used to train the GPT-4 for polymer chemistry and I think this is actually what Citrine Informatics is attempting to do (happy to talk to the Citrine people). I think for now though we can look to companies like Molydyn to fill our basic modeling needs for those of us looking to dip our toes into the water. 

Molydyn is 100% bootstrapped by the founding team. They are based in Bristol, UK and they hope to make computational chemistry more readily available by using first principles physics. I’d recommend trying them out and you have something out of their model in about 10-15 minutes if it’s simple and maybe overnight if it’s complex. If you have the time I urge you to check them out, play around with designing some structures, and see what it can do. 

They are looking for additional funding so if you are interested in talking to Matt you should email Molydyn at hello@molydyn.com.

Tony

1. Substack doesn’t let me do superscripts. I know, annoying right?