We're excited to launch a new Rowan workflow for subscribers today—macroscopic pKa prediction! We use a retrained Uni-pKa-based model ("Starling") to quickly predict per-microstate free energies, and use these values to compute pKa, pI, logD, & blood–brain-barrier permeability:

pKa prediction is crucial in computer-assisted drug design, but most pKa-prediction methods focus only on a single microstate. This can work well in simple cases, but many druglike molecules have dozens or hundreds of relevant microstates! (@jonwzheng has written about this)

We were impressed by the Uni-pKa work from DP Technologies, which uses a neural network to predict per-microstate free energies, but wanted to do more than just win pKa benchmarks. So we retrained a smaller Uni-pKa model internally, which we call Starling. It's quite good:

Once we have microstate populations at each pH, we can easily figure out the isoelectric point of arbitrary small molecules, compute logD from per-microstate clogP values, and more.

We can use Starling & AIMNet2 to quickly estimate state-corrected solvation energy, which can be used to predict Kp,uu (the unbound brain-to-plasma partition coefficient). This gives us a physics-based way to compute blood–brain-barrier penetrance with good accuracy:

If you want to learn more, check out our preprint. And if you want to use Rowan's macroscopic pKa workflow in your organization or company, feel free to get in touch!

Here I share an XGBoost model that delivers a 25% CAGR with minimal drawdown on Visa stock. In this free Substack post I share code and commentary for a powerful Machine Learning strategy that delivers powerful returns.

@kulesatony this is cool!

@kulesatony thanks!

fascinating work sir! exciting to see more advancements in drug discovery tools.

This looks awesome!

thanks!