My lab studies the folding of proteins via structural bioinformatics, molecular simulations, and protein design. We are focused on the folding of green fluorescent protein and its application as a computationally programmable peptide biosensor. We use Hidden Markov models, kinetic simulations, data mining, machine learning, whole gene assembly and other molecular biology techniques, and Xray crystallography.

Bioinformatics techniques were developed towards the grand goal of ab initio protein structure prediction from sequence, using a novel Hidden Markov Model for protein local structure motifs called HMMSTR. Spinning off from HMMSTR were algorithms and servers for predicting contact maps, fully-automated ab initio structure prediction (a Faculty of 1000 paper), for non-sequential alignment of non-homolog structures (SCALI and FlexSNAP), and for improved pairwise sequence alignment of distant homologs (HMMSUM). Also, a novel molecular dyanmics algorithm was developed called CALF, that uses knowledge-based potentials and a greatly simplified representation of the chain. Torsion space equations of motion, developed in the lab, were used in this dynamics program (cover photo in Protein Engineering). A fast calculator of molecular surfaces was developed (MASKER) using boolean masks, which has since been incorporated into more recent software.

Theoretical studies of the determinants of protein folding pathways have been a central focus of the lab. We developed a rule-based model for building contact maps from sequences using folding pathway-based rules, and informatics-based pairwise residue-residue contact energies. This work was followed by a graph-based model for unfolding proteins, called UNFOLD. An improved version of that model was developed that used a detailed atomistic representation of the protein structure and geometric unfolding moves, called GEOFOLD, that predicts kinetic stability and the effect of disulfide bond engineering. CALF simulations were used to explain how local structure formation changes the shape of the folding funnel. Bioinformatics and theoretical studies proved for the first time that the formation of helices in proteins generates a transient torque, which explains the predominant right-handedness of helical crossover structures, dubbed the "Phone Cord Effect." Structural bioinformatics of psychrophilies revealed a new hypothesis for how water-filled cavities may increase enzyme activity at low temperatures. Structural bioinformatics and theretical folding work continues as a minor thrust in the lab.

Protein design is the new major thrust in the Bystroff lab. A protein design program (DEEdesign) has been deveoped that incorporates new energy functions and design algorithms, including "piecemeal" design to enable large, distributed protein design tasks with small resources, and "plastic" design to model proteins as moving rather than fixed targets. Energy terms have been optimized using machine learning, exploring novel hydrogen bonding interactions, solvation terms, and covariance between energy terms. Recently, the lab has switched RosettaDesign, a protein design tool developed by team programming by Rosetta Commons. The lab runs both protein design packages on the CCNI supercluster through a webserver.

Experimental efforts in the lab -- gene construction, cloning and screening -- are directed by Prof. Donna Crone. The lab has developed the "leave-one-out" concept of protein reconsitution. If part of the protein is omitted and it is a part that folds late in the folding pathway, then the leave-one-out or LOO protein will have a strong affinity for a peptide having the sequence of the missing piece. We have explored various versions of LOO green fluorescent protein (LOO-GFP). GFP has been "re-wired" by synthesizing a permuted sequence, and was found to fold and glow. It has been mutated to fold faster, and disulfides have been introduced in two places. Designed or engineered GFPs are synethesized using assembly PCR, an may be screened using a high-throughput protein complementaion assay that has been refined in the lab.

Combining computational protein design and experimental high-throughput sequencing is the current major thrust in the lab. We are computationally designing LOO-GFP to detect peptides from dengue virus proteins. We are experimentally screening large libraries of computationally designed seqeunces, purifying and characterizing the biosensors, and solving crystal structures.

Computational modeling using system dynamic can predict the outcome of a well-described dynamic system, such as human population on Earth. To date, population models have been demographic models, considering only birth rates, death rates and migration. But the future of the human population depends on how much we affect the environment. Climate change and other anthropogenic changes to the Earth's ecosphere will affect the carrying capacity for humanity. Where will it lead? World4 is a system dynamics model for human population.