Abstract

Accurate prediction of protein functions solely from its amino acid sequence is 7 of paramount importance, particularly in the development of new drugs. An 8 ARTMAP neural network (NN) is employed to predict a protein’s function 9 based only on its amino-acid (AA) sequence. For our protein database, a Gene 10 Ontology-based search against the UniProt/SwissProt database for “DNA se-11 quence-specific binding proteins”. The search complement set was also re-12 trieved. For training and testing, various size datasets were generated. Datasets 13 were generated either by random sampling from the existing categories or by 14 classifying the proteins first into sub-groups based on a similarity measure and 15 then randomly sampling from each sub-group. Our NN’s performance with the 16 latter method performed better than with the former method in every size da-17 taset. Our NN has been successful in predicting the function of a protein from its 18 AA sequence by extracting a shared sequence-specific feature that is linked to 19 specific DNA binding proteins. This result is of major importance in structural 20 biology and biomedicine as it can provide a basis of the development of highly 21 specific tools for genome modification and gene therapy.