Artificial peptides bear a great potential for the development of peptidic and peptidomimetic drugs. A computer-based strategy for searching sequence space for desired amino acid sequences employing artificial neural networks and evolutionary algorithms has been developed. The method is thought to reduce the screening effort needed by reducing the number of bench experiments and in vitro tests. A quality function is represented by a neural network that has been trained on the prediction of functional values for peptides. This quality function is used to divide the sequence space into regions of higher and lower quality, and an evolution strategy can be used to efficiently exploit sequence space. Based on this technique signal peptidase substrates and antibody-binding peptides revealing the desired functions have been identified.
Dr. Gisbert Schneider
Free University of Berlin
Institute for Medical/Technical Physics and Laser-Medicine
Berlin, Germany
