The rational design of amino acid sequences by artificial neural networks and simulated molecular evolution: de novo design of an idealized leader peptidase cleavage site

Abstract

A method for the rational design of locally encoded amino acid sequence features using artificial neural networks and a technique for simulating molecular evolution has been developed. De novo in machine design of Escherichia coli leader peptidase (SP1) cleavage sites serves as an example application. A modular neural network system that employs sequence descriptions in terms of physicochemical properties has been trained on the recognition of characteristic cleavage site features. It is used for sequence qualification in the design cycle, representing the sequence fitness function. Starting from a random sequence several cleavage site sequences were generated by a simulated molecular evolution technique. It is based on a simple genetic algorithm that takes the quality values calculated by the artificial neural network as a heuristic for inductive sequence optimization. Simulated in vivo mutation and selection allows the identification of predominant sequence positions in Escherichia coli signal peptide cleavage site regions (positions -2 and -6). Various amino acid distance maps are used to define metrics for the step size of mutations. Position-specific mutability values indicate sequence positions exposed to high or low selection pressure in the simulations. The use of several distance maps leads to different courses of optimization and to various idealized sequences. It is concluded that amino acid distances are context dependent. Furthermore, a method for identification of local optima during sequence optimization is presented.