Predicting Solubility, Google Sign-in, and User Spotlights
different approaches to solubility prediction; Rowan's solubility workflow; sign in with Google, vox populi vox dei; a chance to be featured on our blog
At Rowan, we’ve been focused on trying to bring the best of computational chemistry—a field that’s dominated by dedicated practitioners—to bear on the problems of experimental chemists.
We’re working towards a future where every scientist has access to the best modeling tools available for the problems they face, whether physics-based or machine-learned. Today, we’re excited to share two updates as we move towards that vision: we’ve built a tool around an impressive new solubility prediction model, and we’re adding support for “Sign in with Google.” Plus, we’re excited to offer our users a chance to share their work with our community—read on to learn more.
Do Machine-Learned (ML) Solubility Models Work?
Traditionally, solubility has been predicted using either the Hildebrand solubility parameter or Hansen solubility parameters. The Hildebrand solubility parameter, δ, depends on the energy need to vaporize a molecule, and the three Hansen solubility parameters—dispersion, dipolar interaction, and hydrogen bonding—attempt to correct shortcomings of the Hildebrand parameter. Both models, however, are very low-dimensional and depend on costly real-world measurements of the parameters for their predictions.
Machine-learned models, on the other hand, can learn solubility trends from large data and, despite lacking the elegant theoretical simplicity of a minimal-parameter model, learn trends quite well. Recently, Lucas Attia, Jackson Burns, and co-workers at MIT released fastsolv, a pre-trained model for organic solubility prediction (web interface, GitHub). This model was trained on 54,273 solubility measurements and, as the title of their paper reports, approaches the limit of aleatoric uncertainty (the intrinsic uncertainty of the data).
We were impressed by their efforts and want to help make this model widely available! We encourage the interested reader to see Jonathon’s new blog post on the evolution of solubility prediction methods.
Fastsolv on Rowan for Solubility Prediction
We’ve worked to turn the fastsolv model into a tool built to help with the problem of solvent selection. Chemists are constantly in the situation where they have a solute in hand and are forced to rely on their chemical intuition to guide their choice of solvent.
Rowan’s new multi-solvent temperature-dependent solubility predictor makes inputting a solute, selecting solvents, and analyzing the predicted solubilities frictionless for experimentalists facing a solvent selection decision.
fastsolv is built around the SMILES molecular representation, and we’ve built our interface to support this: you can input solutes from PubChem or via SMILES input with real-time 2D structure visualization, quickly select solvents from a pre-defined list or add your own via SMILES, and all the predictions run in under a minute.
Results are plotted on a responsive temperature-dependent solubility chart. Predicted solubilities are plotted for each solvent, and shaded uncertainties are shown around each line. Clicking on the graph’s legend will toggle the visibility of each line, and all of the results can be downloaded as a CSV if needed.
Sign In with Google
One thing we’ve heard a lot from potential Rowan users since we launched our web platform is that they don’t want to worry about creating a separate account just for Rowan. Our mission is to put the best computational tools out there in the hands of scientists, and we want to make this as easy as possible—so we’re very excited to be announcing support for “Sign in with Google.” The name pretty much says it all:
If your account is already associated with a Google-managed email address and you’ve verified that email, “Sign in with Google” will automatically link to your existing Rowan account.
User Spotlights
As Rowan’s community grows, we’ve gotten the chance to meet more and more people doing all sorts of impactful and exciting research. We always enjoy talking with and learning from our users—and after months of having these conversations, we realized we’d love to give some of our users a chance to share their research with our audience too.
Over the next few months, we’re hoping to write a handful of short blog posts highlighting members of our community, what they’re working on, and how computation has impacted their work. If you’re a current Rowan user and you’re interested in sharing your research with a wider audience, please let us know by filling out this Google Form and we’ll be in touch with you!