Protein folds and their recognition from sequence

Abstract

Abstract Until relatively recently there had been little progress in methods for protein structure prediction (1). Secondary structure prediction had hardly changed since the early 1970s with little improvement in accuracy despite much work and an increase in the size of the database used to derive the parameters (2). Methods for tertiary structure prediction were hampered by the relatively low accuracy of secondary structure prediction. The latter (3–7) were predominantly knowledge-based, in that most methods relied on secondary structure propensities derived from proteins of known structure. As the number of structures rapidly increased during the 1980s, it became apparent that there are also rules which govern the arrangement of secondary structures into their globular fold. For example the connections in ²±² units are always right-handed (8,9), and helices and sheets tend to pack in preferred orientations (10–12). The challenge at the end of the 1980s was how to use this information to improve tertiary structure prediction.