Computational Protein Design as a Cost Function Network Optimization Problem

Abstract

Abstract. Proteins are chains of simple molecules called amino acids. The three-dimensional shape of a protein and its amino acid composition define its biological function. Over millions of years, living organisms have evolved and produced a large catalog of proteins. By exploring the space of possible amino-acid sequences, protein engineering aims at similarly designing tailored proteins with specific desirable properties. In Computational Protein Design (CPD), the challenge of identifying a protein that performs a given task is defined as the combinatorial optimization problem of a complex energy function over amino acid sequences. In this paper, we introduce the CPD problem and some of the main approaches that have been used to solve it. We then show how this problem directly reduces to Cost Function Network (CFN) and 0/1LP optimization problems. We construct different real CPD instances to evaluate CFN and 0/1LP algorithms as implemented in the toulbar2 and cplex solvers. We observe that CFN algorithms bring important speedups compared to the CPD platform osprey but also to cplex.

iCFN: an efficient exact algorithm for multistate protein design

“…With the generic formulation given, we proceed to introduce our exact algorithms based on CFNs. CFN is the state-of-the-art approach to single-state protein design with a single substate ( Allouche et al. , 2012 ; Traoré et al.…”

“…Recently, a new framework of exact algorithms called cost function network (CFN) has been introduced to re-formulate single-state design as a weighted constraint satisfaction problem (WCSP) ( Larrosa, 2002 ; Schiex et al. , 1995 ) modeled through a CFN and to solve it using depth-first branch-and-bound (DFBB) ( Allouche et al. , 2012 ; Traoré et al.…”

Dead-End Elimination for Weighted CSP

“…for spot5 using ternary cost functions). The celar [4] (n ≤ 458, d ≤ 44) and computational protein design [1] (n ≤ 55, d ≤ 148) have been selected as they offer good opportunities for neighborhood substitutability, at least in preprocessing as shown in [14,20]. We added Max SAT combinatorial auctions using the CATS generator [27] with 60 goods and a varied number of bids from 70 to 200 (100 to 230 for regions) [23].…”

“…In addition, we solved the celar7-sub1 instance with the same max degree ordering: EDAC+DEE r solved in (7.7 seconds, 57,584 nodes, 0.96 removals per node), and EDAC+PSNS r in (69.5, 39,346, 7.2), or (86.4, 70,896, 6) as reported in [26]. Secondly, we used a dynamic variable ordering combining Weighted Degree with Last Conflict [25] and an initial Limited Discrepancy Search (LDS) phase [18] with a maximum discrepancy of 2 (option -l=2, except for protein using also -sortd -d: as in [1]). This greatly improved the results for all the methods and benchmarks except for warehouse where LDS slowed down the methods.…”

“…In particular, soft neighborhood substitutability (SNS) [3,26] allows us to detect dominated values in polynomial time under specific conditions for Weighted Constraint Satisfaction Problems (WCSP). In a different community, similar dominance rules and others, called dead-end elimination (DEE) criteria, have been studied for many years in the context of computational protein design [1,13,14,16,17,28,30]. However to the best of our knowledge, these criteria have never been used during search, possibly due to their high computational cost.…”