Research

Integrating Structural Biology with Systems Biology

Some time ago we proposed that protein structure space is not adequately utilized in the detection of functional and evolutionary relationships between proteins. We have been developing methods that use geometric similarity to search broadly for novel functional relationships throughout protein-structure space using what we call the “Structural Blast” concept to identify small structural fragments of proteins that reveal functional information. Our most intense focus has been on the prediction of protein-protein interactions (PPIs) on a genome wide scale. Our PrePPI (Predicting Protein Protein Interactions) algorithm uses the Structural Blast concept to predict interactions, for the entire 200 million possible PPIs in the human proteome. Over a million high confidence PPIs appear in our public database along with structural models and functional information for each interaction. No other structure prediction method can function on this scale. A derivative of PrePPI, P-HIPSTer (Pathogen-Host Interactome Prediction using Structure similarity) was used to predict PPIs for virus-human interactions for 1000 human infecting viruses. In another application, PrePPI was used together with context-dependent interaction networks to produce tumor- and tissue-specific signaling networks.

We are currently improving and extending PrePPI and P-HIPSTer by integrating AlphaFold structures and a variety of machine learning methods into their computational pipelines. We are using these to advances to predict novel signaling networks for multiple biological processes, particular those related to cancer and to pathogen infection. We have also used the Structural Blast concept to develop a database of Protein Compound Interactions (PrePCI). PrePCI provides predictions for billions of potential PCIs thus allowing us, for example, to suggest approaches to find multiple druggable targets in PrePPI and P-HIPSTer-predicted networks.

Protein-protein interactions and their role in the development of neuronal connections

Our research focuses on the relationship between protein-protein interactions and cell-cell recognition, the ability of cells of a certain type to assemble into larger structures. Our recent interest has been in the role of PPIs in the development of the nervous system. Our most notable achievement has been the elucidation of the mechanism of the barcoding of individual neurons. Each neuron has a unique barcode which is required so that neurites form the same cell body avoid one another whereas neurites from different neurons are able to form synapses. A barcode is necessary to mediate recognition of “self” from “non-self” and it involves a remarkable mechanism based on the formation of ordered lattices of adhesion proteins in neuron-neuron interfaces.  The discovery was made utilizing a combination of biophysical studies, crystallography, CryoElectron Tomography, statistical mechanical theory and multi-scale simulation.

We are currently focusing on the design of adhesion receptors with altered properties so as to study the molecular determinants that mediate the highly specific wiring of neuronal circuits we use computational approaches to design proteins with altered binding specificities, test their properties with surface plasmon resonance (SPR) measurements and then, through collaborations with neuroscience researchers around the world, determine how these alterations affect the development of the nervous system in flys and in mice. Our research program, which combines computational structural biology and developmental neuroscience, is truly unique.

Antibody optimization through protein design

There is great interest in the use of antibodies to combat viral infection. Antibodies are produced in infected patients but are generally not useful as widely used therapeutics. However using these antibodies as starting points, and given a structural model of an antibody-antigen interactions, the hope that more tightly binding antibodies can be designed which will have therapeutic potential. We use a variety of computational techniques to discover stabilizing mutations and then test our predictions using SPR measurements. The work is done in close collaboration with our colleague, Professor Lawrence Shapiro, who is an eminent crystallographer.

The computational method of choice is Free Energy Perturbation simulations (FEP) which exploit extensive molecular dynamics calculations. We collaborate in this effort with Professor Richard Friesner of the Columbia Chemistry Department, a pioneer in the development of FEP methods. The method has been used extensively for optimizing drug protein interactions but our application to the effect of mutations on protein-protein interactions is particularly challenging since the relevant interfaces are quite large and often flexible. We have made exciting progress on this difficult problem and our research effort to design antibodies for a variety or viral pathogens is ongoing.