Computational de novo protein design with flexible templates

De novo peptide or protein design starts with a flexible three-dimensional protein backbone template and involves the search for all amino acid sequences that will fold into such a template. The sequences will vary in their abilities to perform the same function as that of the mother sequence. In the usual case, we are interested in the sequences that have higher activities than the native protein. The problem is of paramount importance to protein engineering and peptide and protein drug discovery.;The de novo design problem can be posed as two questions: (i) "How should a flexible design backbone template be defined?"; and (ii) "How can the sequence search on such a flexible design template be performed efficiently to design proteins of practical sizes?". To answer the first question, all current approaches for constructing a flexible design template are studied. It is concluded that the best method is through a continuum backbone and continuous ranges of dihedral angles. In such a truly flexible design template, all inter-atomic distances and bond angles are semi-infinite sets bounded between upper and lower bounds.;Regarding the second question, work is done based on improving the original de novo design framework proposed by Klepeis et al. (2003)(Klepeis et al., 2004). This framework designs proteins through two stages: (i) a sequence selection stage which predicts low energy sequences compatible with the target fold using a novel mixed-integer programming model; and (ii) a fold validation stage which calculates the probability of each sequence from the previous stage to fold into the flexible design template, using a global optimization approach under a full-atomistic force field. Computational efficiency of the framework is improved by formulating a different optimization model for the sequence selection stage, and performing an approximate method for fold validation via a protocol for NMR structure refinement instead of the rigorous but computationally expensive global optimzation approach. To incorporate backbone flexibility more explicitly into sequence selection, two new integer programming models are also developed to handle flexible design templates with multiple structures.;The new framework is applied to five de novo designs: (i) de novo design of compstatin variants; (ii) full-sequence design of human beta-defensin-2; (iii) design of inhibitors to HIV-1 gp 41 and gp 120; (iv) redesign of the carboxy-terminus of complement 3a; and (v) design of antagonists to complement 5a receptor. These examples demonstrate how the issues of defining a flexible design template, deriving a mutation set, imposing useful constraints to improve the quality of the solutions, and performing analysis of the predicted sequences can be solved when using the framework.;The thesis also outlines an independent chapter which presents a novel mixed integer nonlinear programming model on designing a biosensor of receptors for optimally recognizing a known set of ligands.