Influenza is a deadly virus that remains a worldwide problem despite vaccinations, modern medicine, and public awareness. After infection, antiviral drugs are the primary treatment option. One common antiviral strategy is to prevent viral entry by targeting the hemagglutinin protein (HA) which sits on the surface of the virus. There are currently 18 known subtypes of HA, resulting in the efficacy of many antiviral drugs being limited. The goal of this project is to computationally design disulfide-closed cyclic peptides that target conserved binding regions on HA and test for broad spectrum activity against multiple HA subtypes. To start, I used RFdiffusion to generate peptides that are docked against the H1 subtype of HA. Then I used ProteinMPNN to optimize the amino acid sequence of the peptides. Afterwards, I used a custom written program named Protein Transform to take the peptides and dock them on the different HA subtypes. I then ran a custom program that uses RoseTTAFold2 (RF2) to predict the root mean square error (rmsd) between the target binding spot and the spot RF2 believes the peptide will stay and score the interface predicted align error (ipae), which is a confidence metric for the binding interaction. Next, I ran a Rosetta program to calculate the interaction energy (ddG) between the protein-peptide complex. The designs were then filtered based on ipae, rmsd, and ddG scores using our lab’s standard filtering thresholds. Our lab then used yeast surface display screening to test the peptides and measure the binding affinity (KD) against the H1 subtype. Then we verified the hits using surface plasmon resonance. We anticipate to see KD values in the range of 10 to 1000 nM. Identifying a lead compound with broad spectrum HA binding is an important step to creating a drug to treat any strain of Influenza.

B-cell lymphoma-extra large (Bcl-xl) is a mitochondrial transmembrane protein that acts as an anti-apoptotic protein by sequestering the apoptosis-inducing proteins Bim, Bak, and Bad. This prevents the release of cytochrome c from the mitochondria, preventing activation of apoptosis pathways. Higher levels of Bcl-xl expression are commonly found in cancer cells. This contributes to the prevention of apoptosis in cancer cells, allowing them to proliferate uncontrollably. Bcl-xl is an incredibly important target for cancer therapeutics. A Bcl-xl binder would inhibit the interaction between Bcl-xl and apoptosis inducing proteins, allowing cancer cells to undergo apoptosis. In my research, I am using deep learning methods to design cyclic peptides that bind to Bcl-xl. To design the binders, I used RFDiffusion - a generative diffusion model - to produce thousands of cyclic peptide binder scaffolds bound to Bcl-xl. I then used a sequence-based deep learning tool to generate multiple sequences for each backbone design. The resulting binders were computationally validated with the highly accurate, machine-learning-based structure prediction tools AlphaFold and RoseTTAFold. Of the 40000 generated cyclic peptides, 2052 were predicted to bind to Bcl-xl based on standard metrics. Along with their excellent metrics, these designs show a high structural similarity and binding location to the known Bcl-xl binders Bim, Bak, and Bad. The designs were clustered by backbone into 350 unique clusters. We synthesized the top designs and identified which peptides display binding to Bcl-xl through a Homogeneous Time-Resolved Fluorescence (HTRF) assay. A successful Bcl-xl binder has the potential to serve as the basis for an effective and affordable cancer therapy.