"A combination of computational and experimental techniques to design binders against P53"

The intrinsic region of p53 binds to various proteins, each with distinct functions. Our goal is to design inhibitors for three of these interactions by targeting the protein partners instead of p53 itself. The best designs, selected from 96 wells, will then be analyzed using multiple methods

"Predictive modeling for pMHC targets by TRACeRs"

We developed a novel platform, called TRACeR, to target the peptide antigen in Major Histocompatibility Complexes (MHCs) using only a small number of active site residues to define the specificity. Highly specific MHC binders are currently discovered through high throughput library selection, but we hope to annotate the binder-to-target sequence space to be able to create MHC binders for any given peptide sequence without wet-lab selection. To this end, this project will involve (1) experimental data collection on binder:target pairings, (2) building a computational model to generate designs, (3) followed by testing of designs on unseen targets.

"Mapping and designing protein-peptide interaction specificity landscapes using AI, biophysics and deep sequencing"

The C-terminal floppy segments of proteins in the human proteome provide a sequence-unique Achilles heel to selectively target any protein inside cells. Naturally occuring proteins like PDZ domains are known to bind to C-terminal segments but with weak affinity and selectivity. We are developing CODACs - Computationally Designed Affinity Clamps - that can bind target C-terminal peptides with high affinity and specificity, using both AI-based (RFDiffusion, MPNN) and biophysical simulation-based design tools. Experimental characterization of designs involves measuring binding and specificity landscapes using high-throughput display experiments and deep sequencing.

" Computational design of de novo proteins to control biological signaling"

We are working towards engineering synthetic signaling systems built from de novo designed protein components that can recognize inputs, transduce signals, and control programmable outputs. We have a range of projects to create proteins with custom-designed shapes to recognize specific signals and to engineer switchable protein structures. We integrate computational design, including recent advances from deep learning, and experimental characterization in vitro and in cellular systems.

"AI methods for protein structure prediction"

Knowledge of protein structure is paramount to our understanding of biological function and for developing new therapeutics. Deep-learning structure modeling techniques such as AlphaFold have shown great success but still exhibit notable limitations for numerous types of proteins, such as dynamic proteins with multiple conformations, orphan proteins that have no close homologs, or intrinsically disordered proteins. One of the most promising avenues to overcoming these remaining limitations is integrating experimental data directly into AlphaFold2. We are developing integrative deep-learning based modeling techniques, that enable prediction of protein structure from sparse experimental data. We are particularly interested in using mass spectrometry and deep mutational scanning data for predictions.

"Generative AI for drug selectivity"

A broad challenge in drug discovery is finding compounds that selectively target certain receptors or conformational states. State of the art molecular simulations dock ultra-large chemical libraries at multiple sites and pick compounds based on predicted differential binding. Recently, generative AI docking methods have shown promise at conditional ligand and pose generation tasks. In this project, you will leverage experimental data capturing relative activity to guide and fine-tune generative AI methods to make them better at predicting selective compounds and then benchmark against state-of-the art molecular simulation approaches. As a case study you will prospectively predict compounds to activate KCNQ ion channels to treat epilepsy, tinnitus, and other diseases of neuronal hyperexcitability.

"Applying high-throughput experimental data to guide computational protein design"

Today, most computational protein design tools like Rosetta use the features of natural protein structures (which amino acids like to be near each other, what types of structures are very common, etc.) to guide the design of new proteins. However, for many applications, we want to design proteins with properties far beyond what already exists in nature. To achieve this, we need new sources of data - not just natural protein structures - that can guide design into new territory. Our lab develops new experimental methods to measure properties like folding stability, aggregation, and dynamics for tens to hundreds of thousands of designed or natural proteins at the same time. We then use these new large datasets to guide protein design proteins. We have a range of different projects focused on basic science, therapeutic development, and tools for synthetic biology. Each person's project is described on our website (www.rocklinlab.org). We will work with an intern or post-bac to find which project in our lab is best for their interests.

"Reshaping protein energy landscapes to optimize dynamics/function"

The Smith lab is currently developing and applying methods for combining Rosetta design calculations with molecular dynamics simulations to reshape the energy landscapes of proteins. We are applying these techniques to several unique systems: 1) mini Flurorescence Activating Proteins (mFAPs) are beta barrels de novo designed in Rosetta to bind and activate the fluorescence of small molecule chromophores. Our lab developed an algorithm that can for the first time predict how mutations to the protein affect chromophore dynamics and thus brightness. We are reshaping the energy landscape of these proteins to favor macrostates with high protein function. 2) Rosetta-designed miniprotein binders potentailly represent a new class of protein drugs that can be readily created using the structure of the target protein alone. However, we have found that model miniproteins exhibit unexpected and undesired conformational heterogeneity and spontaneous postranslational modifications. We are redesigning these proteins to avoid such deleterious outcomes. You will learn to use Rosetta with free energy data analysis techniques, optimize algorithm parameters with existing datasets, and apply these techniques to improve protein function.

"Protein design for new protein click chemistries"

Scalable routes for site-specific, defined immobilization of proteins on supports is important for biotechnology. For regenerative medicine and drug delivery, tethering proteins to hydrogels is essential for controlled release and delivery in vivo. Conjugating drugs to antibodies allows for precise targeting of chemotherapies in cancer. For sustainable processes involving enzymes or affinity reagents, proteins are often attached to supports in packed columns. Wbile there has been >60 years of development of chemistries that link proteins to supports, there are gaps in scalable manufacturing routes for site specific immobilization. Mutations to cysteine or non canonical amino acids suffer from yield and downstream purification liabilities. This project seeks to use protein design to learn new short peptide sequences which allow for site-specific conjugations with epoxy or other suitable electrophilic functional groups. We will use generative AI to design structures and sequences, and test conjugation in vivo using prokaryotic and eukaryotic expression systems. The ideal trainee would be someone with interest in kinetics, protein design, applied biotechnology, and with some chemistry background.