Building hydration-aware AI for protein design
Protein Design Solutions (PDS) is building hydration-aware AI for protein design. We reveal how water responds to protein surfaces—in context—showing physical behavior that sequence and static structure alone can miss.
Hydrophobicity is not a fixed property of an amino acid.
It depends on the residue’s surrounding chemistry, surface geometry, neighboring groups, and the collective response of water.
What standard tools see
Conventional descriptors often estimate surface hydrophobicity from amino-acid identity, solvent exposure, and static structural features.
The same amino acid is treated as having approximately the same hydrophobic tendency wherever it appears.
What PDS sees
PDS predicts how the surrounding water responds to each protein-surface region in its actual three-dimensional environment.
Local chemistry, shape, and neighboring residues can cause the same amino acid to behave very differently.
Future verified figure: same protein, same molecular orientation, and one annotated region whose interpretation changes when hydration context is included.
Same amino acid. Different neighborhood. Different hydration.
The relevant signal is not only in the protein. It is in the surrounding water.
Protein designers routinely navigate sequence, structure, affinity, charge, and other measurable properties. But another variable shapes how a protein behaves: how water responds to each surface region in context.
PDS makes that coordinate visible—and usable for design.
Visible design space
Expanded design space
PDS helps teams decide which related candidates deserve experimental resources by adding context-dependent hydration information to protein-engineering workflows.
Rank closely related antibody and protein variants before experimental testing.
Identify context-dependent hydration patterns that conventional amino-acid scales can miss.
Add interpretable hydration information to existing protein-engineering workflows.
PDS currently focuses on a commercially common question: given one parent protein and many closely related variants, which candidates should a team prioritize for laboratory testing?
One parent
A known antibody or protein starting point.
Related variants
Variants proposed through affinity maturation, protein engineering, or computational design.
PDS hydration analysis
Hydration-derived descriptors and residue-level maps.
Ranked panel
Selected candidates proceed to physical production and testing.
The same hydration information can ultimately help guide which variants to create—not only which proposed variants to select.
From expensive water physics to hydration-aware AI
PDS uses custom enhanced-sampling molecular simulations and statistical thermodynamics to measure hydration signals that cannot be inferred reliably from sequence or a static structure alone. Protein hydration depends in nontrivial ways on local chemistry and surface topography, and its most informative signatures appear in rare, collective low-density fluctuations of water.
The underlying approach was pioneered by PDS founder Nicholas Rego, PhD, and collaborators in Prof. Amish Patel’s lab at the University of Pennsylvania. Through a completed NSF SBIR Phase I, PDS extended that work into a proprietary, first-of-its-kind dataset of residue-level protein hydration behavior and trained fast models that make those expensive calculations usable at protein-design scale.
These hydration-aware models produce interpretable, residue-level signals that can help scientists rank variants, understand why closely related proteins behave differently, and design better-behaved proteins.
01
Protein in explicit water.
02
Context-dependent, residue-level response.
03
Expensive calculations distilled into rapid prediction.
04
Variant ranking plus hydration maps.
Hydration begins as an expensive physical calculation. PDS turns it into a design-scale coordinate.
Retrospective benchmark
Across 125 parent-excluded variants in four antibody-variant panels, every displayed method discarded the predicted-worst half within each panel.
PDS removed 28 of the 32 variants that experiments identified as the worst-performing quartile.
This result is consistent with hydration providing decision-relevant information that conventional scoring methods do not fully capture.
Equal discard budget within each parent–variant panel
Among the 63 retained candidates, PDS left 4 worst-quartile variants, compared with 10 for Aggrescan3D and 11 for DeepSP.
Retrospective analysis: experiments defined which variants were worst; each method predicted which half of every panel to remove.
The underlying hydration framework can be applied to protein structures and supplied conformational ensembles across protein classes. Current commercial validation is strongest in antibody Fv/scFv parent–variant panels.
PDS’s current specialty is context-dependent protein hydration. Over time, the same simulation-to-model strategy can expose additional physical coordinates for protein design.
Nicholas Rego, PhD, founded Protein Design Solutions to translate fundamental protein biophysics into practical tools for protein design. He developed the core hydration methodology during his doctoral and postdoctoral research at the University of Pennsylvania with Prof. Amish Patel, then led its translation through NSF I-Corps and a completed $275,000 NSF SBIR Phase I.
PDS is seeking biologics and protein-engineering teams interested in hydration-aware variant ranking and validation studies.
We are also interested in collaborators exploring hydration-aware protein design beyond antibodies.