The prediction of protein adsorption to surfaces from solution is a perennial unsolved problem in biomedicine, physical chemistry and other fields. Here we used neural networks and the previously developed Biomolecular Adsorption Database (BAD) to predict the amount of protein adsorbed by a set of five descriptors of the protein, surface and solution. We find a moderately good predictive ability if very large adsorption values are present and a good fit if these few outliers are eliminated. With a growing number of entries in the BAD, we expect the accuracy of the predicted values to increase substantially. This paper presents for the first time a universal and stand-alone quantitative predictor of protein adsorption.

Date Published: 2 January 2008
PDF: 6 pages
Proc. SPIE 6799, BioMEMS and Nanotechnology III, 679911 (2 January 2008); doi: 10.1117/12.768952

Published in SPIE Proceedings Vol. 6799:
BioMEMS and Nanotechnology III
Dan V. Nicolau; Derek Abbott; Kourosh Kalantar-Zadeh; Tiziana Di Matteo; Sergey M. Bezrukov, Editor(s)