SAPT, Protein Preparation, and Starling-Based Microscopic pKa
interaction energy decomposition w/ SAPT0 & a warning; making protein preparation more granular; catching forcefield errors earlier; microscopic pKa via Starling; internship applications now open
This week at Rowan, we’re releasing a new workflow (interaction energy decomposition), some changes to how we handle protein preparation and submission, and a new Starling-powered microscopic-pKa method.
We’ve also released an internship interest form for summer 2026 and an FAQ page for credit-related questions. (Early versions of this newsletter had a broken link here; sorry! This is now fixed.)
Interaction Energy Decomposition
We’re releasing a new interaction-energy-decomposition workflow that uses symmetry-adapted perturbation theory (SAPT) to explain the physical origins of intermolecular interactions. SAPT works by using perturbation theory to decompose the interaction energy into various terms, which can then be interpreted to help improve the activity of ligands or catalysts.
Simple energy-decomposition methods use the energy difference between the complex and the fully separated systems to determine the interaction energy, but this is prone to basis-set superposition error and cannot give more fine grained insight onto the nature of the interaction. SAPT builds up the interaction perturbatively from the individual monomers.
In SAPT0 (the simplest level of SAPT theory), monomer wavefunctions are constructed using Hartree–Fock (HF) theory. The unperturbed (zeroth-order) wavefunction for the dimer is formed from the product of the HF determinants and antisymmetrized (symmetry adapted) to ensure the Pauli principle holds.
The SAPT0 interaction energy is then determined via second-order perturbation theory, with terms divided as:
Electrostatic - attraction/repulsion between partial charges on each system
Exchange - Pauli exchange due to antisymmetrization (a purely quantum effect)
Induction - polarization of each monomer by the electric field of the other (includes exchange-induction)
Dispersion - electron correlation
Electrostatic and exchange terms enter at first-order, dominating the total energy, but also mostly cancelling for neutral non-polar molecules. Induction and dispersion enter at second-order and contribute small attractive effects.
For the simple water dimer we obtain a total interaction energy of –4.8 kcal/mol with SAPT0/jun-cc-pVDZ (negative indicates attraction). We can break this down:
There is a strong electrostatic attraction due to the partially positive hydrogen atom being attracted to the partially negative oxygen on the other molecule.
The exchange interaction is positive due to the Pauli exclusion principle preventing same-spin electrons from occupying the same space.
The dispersion is attractive due to the electron correlation that happens between the monomers.
The induction is also attractive, due to the electric field in each monomer inducing polarization in the other monomer, reducing the overall energy compared to the non-polarized case.
Can these abstract quantum-mechanical considerations be useful in real life? While insight is tough to quantify, a collaboration between the University of Southampton and Boehringer Ingelheim found that SAPT performed well on complexes relevant to protein–ligand interactions:
SAPT is seen to give chemically sensible results and arguably provides a more intuitive decomposition than the other [energy-decomposition] schemes…. However, relating the theoretical processes of this scheme and their chemical equivalents can at times be more conceptually complicated than for the variational based schemes.
However, some caution is warranted when using SAPT (or any other energy-decomposition-analysis method). The precise distinctions between the various SAPT energy terms don’t always cleanly map into the conceptual framework of organic chemistry, and it’s possible for inexperienced users to give themselves a false sense of clarity. Corin’s written about this topic previously:
Chemistry is complicated, and ground- or transition-state structures arise from a delicate equilibrium between opposing factors: steric repulsion, electrostatic attraction, bond distances, torsional strain, dispersion, &c…
The ability to break intermolecular interactions down into different components is certainly useful, and it seems likely that some version of EDA will eventually achieve consensus and emerge as a useful tool. But I think the utility of EDA even in the best case is pretty limited. Quantum chemistry is complicated, and if we think we can break it down into easy-to-digest components and eliminate the full nonlocal majesty of the Schrodinger equation, we’re lying to ourselves (or our experimental collaborators). Compute with caution!
While this advice is true for all computations, in our experience it’s particularly true here.
Here’s an example of SAPT on a less trivial system: the result decomposes a relatively small π–π interaction into sizable contributions from exchange (repulsive) and electrostatic effects & dispersion (attractive).
Interaction-energy-decomposition workflows are now available for all Rowan users.
Protein Preparation
Today, we’re making our protein preparation process much more modular and controllable.
After uploading a protein to any submit page, Rowan will now highlight any discrepancies between the selected protein and what the workflow expects.
Any errors or warnings can be addressed using the improved “Prepare” tool in the protein editor, which now exposes a series of controls that let you choose:
Whether or not unresolved residues are identified and added to the protein
Whether any ions and ligands will be removed
Whether any waters will be removed
Whether hydrogens will be added, what pH they will be added at, and whether their positions will be optimized using OpenMM
The entire preparation process takes roughly 20–30 seconds to complete. (As previously, this is accomplished using PDBFixer.)
If any unresolved residues are appended as unsightly tails, you can now also remove them if desired using the editor’s “Delete selected residues” tool. (Preparation may need to be re-run afterwards to add any missing hydrogens.)
Pre-Check Before MD
It’s frustrating to submit a molecular-dynamics (MD) workflow, wait for a GPU machine, and then watch it fail immediately because the forcefield couldn’t be constructed due to an easily fixable issue.
To help users avoid these issues and “fail fast,” we’ve added a pre-check step to every MD workflow submission page. Before launching a pose-analysis MD or protein MD job, you’ll now click a “Check” button.
This pre-check quickly attempts to build a forcefield on an always-on machine. The check will surface any issues encountered so you can fix them before submission.
Starling-Based Microscopic pKa
Last year we published Starling, a retrained version of DP Technology’s Uni-pKa model that predicted microstate free energies directly from SMILES strings. Since then, Starling has been powering our macroscopic pKa workflow—the ability to directly predict free energies allows for very fast macroscopic pKa calculations, particularly when combined with our AIMNet2-based beam-search method.
Today, we’re releasing a version of our microscopic pKa workflow that’s also powered by Starling. For the unfamiliar, microscopic pKa methods only explore the local environment of the input microstate, rather than exhaustively enumerating all microstates like macroscopic pKa methods (you can read more about the difference here).
To run microscopic pKa predictions using Starling, select the “Starling” method on the microscopic pKa prediction page.
This job completed in under 30 seconds and the results can be quickly visualized in Rowan’s GUI. The same checks we run on all pKa predictions will be applied to microscopic pKa predictions run with Starling; here, the workflow warns that there are multiple sites with similar pKa values, which is indeed the case.
While Starling suffers from some known flaws, like attenuated accuracy outside the physiological 5–9 pKa window and poor relative tautomer rankings, we think this method is a useful complement to our existing pKa-prediction offerings.
Summer 2026 Internships
Rowan’s looking to hire 1–2 interns this summer! Interns will get the chance to work alongside our team at a fast-paced scientific startup as we ship new features and scale our company. If this is of interest to you, you’re welcome to fill out our internship interest form (which also has more details on eligibility, expectations, etc).
Credits FAQ
Finally, we often get questions about our credit-based pricing system: how it works, why we’ve structured it this way, and what it means in practice. We’ve released a big credits FAQ that answers all these questions; if you’re confused, check it out here.