Diffusion at the molecular level involves random collisions between particles, the structure of local microscopic environments, and interactions among the molecules involved. Sampling all of these aspects, along with correcting for finite-size effects, can make the calculation of infinitely dilute diffusion coefficients computationally difficult. We present a new approach for estimating the translational diffusion coefficient of biomolecular structures by encapsulating these driving forces of diffusion through piecewise assembly of the component residues of the protein structure. By linking the local chemistry of a solvent-exposed patch of a molecule to its contribution to the overall hydrodynamic radius, an accurate prediction of the computationally and experimentally comparable diffusion coefficients can be constructed following a solvent-excluded surface area calculation. We demonstrate that the resulting predictions for diffusion coefficients from peptides through to protein structures are comparable to explicit molecular simulations and improve on statistical mass-based predictions, which tend to rely on limited training data. As this approach uses the chemical identity of molecular structures, we find that it is able to predict and identify differences in diffusivity for structures that would be indistinguishable by mass information alone.