Research work · 1994

Machine Learning and Biomolecular Modelling

M J E Sternberg, R A Lewis, R D King, S Muggleton

Machine Intelligence 13

Abstract

Abstract Two problems in biomolecular modelling have been tackled by the machine learning program, Golem. The program, based on inductive- based logic programming, takes as input observations and combines them with background knowledge to yield rules. The first system tackled was modelling the quantitative structure-activity relationship (QSAR) of a series of trimethoprim analogues binding to E. coli dihydrofolate reductase. The Golem rules were a better model than standard regression approaches. More importantly, these rules described the chemical properties of the enzyme binding site that were in broad agreement with the crystallographic structure. The other system was the prediction of the secondary structure of proteins to identify regions that form a-helices. On a test set the prediction was 81% accuracy which is better than the results achieved by neural networks. The rules from Golem defined patterns of residues forming α-helices.

DOI: 10.1093/oso/9780198538509.003.0008