Calliper randomization: an artificial neural network based analysis of E. coli ribosome binding sites

Abstract

An artificial neural network based approach has been used in analyzing the translation initiation region of E. coli. The approach is based on using a trained network capable of recognizing a particular region and presenting the network with randomized calliper inputs of the true sequence. The network responds with an error when the regions which have been the main source of knowledge are randomized. Analysis of the E. coli ribosome binding sites using this approach reveal that the initiation codon and the Shine/Dalgarno sequence which are known to be important for translation initiation are also important in imparting knowledge to the network. Further, selectively changing the usually occurring initiation codon AUG, to GUG, UUG and AUU, which occur less frequently, decreases the network performance in accordance with the frequency of their occurrence. This approach can be used as a general method to derive consensus.