Cadre has partnered with Professor Bruce Tidor at MIT to offer their collection of computational tools for improving antibody binding affinity and recycling. These methods reduce experimental time and cost while providing novel solutions to the optimization problem. There are two stages to Bruce's approach. First, a preliminary scan with an efficient branch-and-bound algorithm identifies the most promising antibody variants. Each candidate is then reevaluated using a more computationally demanding scoring function that more accurately models molecular flexibility, electrostatics, and solvent. This computational approach complements more traditional wet-lab directed evolution protocols which have a limited ability to explore sequences significantly different from the wild type. The optimization protocol has achieved up to a 140-fold improvement in binding affinity.

 

- S. Lippow, K. Wittrup, and B. Tidor, Nature Biotechnology, 2007, 25:1171.

Cadre has partnered with Professors Chris Bailey-Kellogg and Karl Griswold at Dartmouth College to offer their suite of computational and experimental tools for producing immunotolerant variants of therapeutic proteins, in which immunogenicity is reduced while therapeutic activity is maintained. Our technology reduces the time and cost of identifying and removing T-cell epitopes; and will be of interest to any company working to mitigate potential anti-therapeutic immune responses. With this technology, therapeutics can more quickly get to market and are less likely to exhibit immunogenicity related failures. The model has been validated on several test systems (including staphylokinase, erythropoietin, and beta-lactamase). (2 page research brief)

 

- Parker, Zheng, Griswold, and Bailey-Kellogg, BMC Bioinf., 2010, 11:180.
- Parker, Griswold, and Bailey-Kellogg, J. Bioinf. Comp. Biol., 2011, 9:207.
- Parker, Griswold, and Bailey-Kellogg, Proc. RECOMB, 2012, 7262:184-198.
- Osipovitch, Parker, Makokha, Desrosiers, Kett, Moise, Bailey-Kellogg, and Griswold, Protein Engineering, Design, and Selection, 2012, (in press).

Cadre has partnered with Professor Christopher Langmead at Carnegie Mellon University to offer their generative statistical model of protein sequences, GREMLIN (Generative Regularized ModeLs of proteINs). Starting with a multiple sequence alignment, the GREMLIN algorithm learns a probability distribution over protein sequences. The model is encoded as a graph and a set of functions. The nodes of the graph correspond to the columns of the MSA and the edges specify the conditional dependencies between the columns. By examining the statistical patterns of sequence conservation and diversity within a protein family, they can gain insights into the constraints that determine structure and function. These insights can be used to design new protein sequences consistent with the protein family and to predict functional activity of variant sequences.

 

- S. Balakrishnan, H. Kamisetty, J. Carbonell, S. Lee, C. Langmead, Proteins, 2011, 79:1061.

Cadre is proud to offer another molecular modeling tool from Christopher Langmead’s lab at Carnegie Mellon University. The GOBLIN (Graphical mOdel for BiomoLecular Interactions) technique performs physics-based free energy calculations for protein-protein and protein-ligand complexes under side-chain, backbone, and ligand flexibility. In other words, GOBLIN employs an efficient algorithm to model both the enthalpy and entropy of molecular interactions. Technically speaking, GOBLIN uses a probabilistic graphical model that exploits conditional independencies in the Boltzmann distribution and employs variational inference techniques to approximate the free energy of binding in only a few minutes. This technology would be useful for researchers studying protein-protein or protein-ligand interactions and the effects of point mutations on binding affinity.

 

- H. Kamisetty, A. Ramanathan, C. Bailey-Kellogg, C. Langmead, Proteins, 2011, 79:444.

A fundamental challenge in drug discovery has been the development of compounds able to avoid or to address the problem of drug resistance. The drug discovery community has demonstrated success in responding to drug resistance by developing second-generation compounds capable of avoiding resistant protein variants after they have been observed in the patient population. We propose an alternate strategy. Cadre Research Labs is developing a suite of computational methods to predict and target drug resistant mutations before they are clinically observed. The approach includes two core components, first a positive screen identifies viable protein variants, then a negative screen identifies those viable mutants with disrupted inhibitor binding. The techniques allow the prioritization of potential drug targets and the planning of optimal drug cocktails. We are putting all the pieces together and should be able to offer this service by Fall 2012.

 

.

Cadre has partnered with Professor Christopher Langmead at Carnegie Mellon University to offer GAMUT (GAmes of MolecUlar conflicT), an innovative new approach for combating drug resistance. GAMUT employs techniques from game theory to anticipate how a molecular target might evolve in response to a given compound and to design game-theoretically optimal drugs and drug cocktails, while retaining specificity for the target. GAMUT has been validated against a number of targets, including HIV-1 protease and PDZ peptide binding. The approach would be useful for those developing optimal treatment regimens involving more than one small-molecule therapeutic.

 

- H. Kamisetty, X. Xing, C. Langmead, International Conference on Machine Learning (ICML), 2011, 28:1153.

Cadre is working with the lab of Professor Bruce Donald at Duke University to develop and apply their structure-based computational methods for protein redesign. Their OSPREY software is specifically designed to identify protein mutants that possess desired target properties (e.g., improved stability, switch of substrate specificity). Most recently OSPREY has been extended to design protein-protein and protein-peptide interactions. The core methods incorporate multiple models of backbone and side-chain flexibility. It implements deterministic search via Dead-End Elimination (DEE) and A* based algorithms and evaluates candidate mutations using both ensemble (K*) and Global Minimum Energy scoring functions. Over the past eight years, these biophysically accurate methods have been validated on numerous biological systems. In addition to extending these techniques, Cadre offers a protein redesign service based on the methods developed in the Donald Lab.

 

- Lilien, Stevens, Anderson, Donald, J Comp. Biol., 2004, 12(6):740-61.
- Georgiev, Lilien, Donald, Bioinformatics, 2006, 22(14):174-83.
- Georgiev, Keedy, Richardson, Richardson, Donald, Bioinformatics, 2008, 24(13):196-204.
- Gainza, Roberts, Donald, PLoS Comp. Biol., 2012, 8(1):e1002335.