Discovery of Protein Structural Constraints in a Deductive Database using Inductive Logic Programming
Ross D. King, Dominic A. Clark, Jack Shirazi, Michael J.E. Sternberg · 1996
Abstract This paper describes a framework for the use of machine learning as a tool to aid scientists in the discovery of patterns in data. The framework is tested by the application of the inductive logic programming (ILP) program GOLEM to the discovery of constraints in the packing of beta-sheets in alpha/beta proteins. These constraints (rules) play a part in the protein folding problem, an important unsolved problem in molecular biology. Constraints were learnt for four features of beta-sheet packing: the winding direction of two sequential sheets, whether two sequential sheets pack parallel or anti-parallel, whether two sheets pack adjacently, and whether a beta-sheet is at an edge. Investigation of the constraints found revealed interesting patterns, some of which were previously known, others that were novel. Novel features include the discovery that the relationship between pairs of sequential strands is in general one of decreasing size, and that more sequential pairs of strands wind in the direction out than the direction in. We conclude that machine learning has a role for scientists as a pattern discovery tool.