Protein Modeling

Our protein modeling team used protein models to inform our wet lab team about ideal locations for residue modifications, and designed novel albumin binders. We used cutting-edge new machine learning tools, such as RFDiffusion and ProteinMPNN, and familiar visualization software, like PyMOL, to build up our protocols and workflows. All of our efforts are described on our model page, with highlights shown here below.

Visualization with PyMOL and AlphaFold Evaluation

We described how we used PyMOL to determine locations for residue modifications for linker reactions and purification tags on the original albumin binding moiety, which we built using AlphaFold2. We additionally used AlphaFold2 to observe and analyze the effects of mutations on the structure and metric values.

RFDiffusion, ProteinMPNN, AlphaFold Workflow

On our model page, we described our methods and workflow for the design of novel protein binders in the section “Computationally Designing and Modeling Novel Albumin Binders with RFDiffusion, ProteinMPNN and AlphaFold.” We contribute all the scripts we wrote, including bash scripts to execute various experiments. We particularly wanted to emphasize the use of these powerful machine learning tools on more accessible computers (no supercomputer needed!), which we think will be useful for future iGEM teams. These tools are increasingly gathering interest in research communities, including for synthetic biology, so we are excited to see how iGEM teams are able to apply these tools and hope our workflow details will be helpful. Additionally, we describe our troubleshooting efforts, the limitations of our models, and methods to increase the speed of these design processes. We have attached a PDF version of our methods section here.

Computational design metrics (with Rosetta and python)

We contribute our metric calculation and evaluation methods. We used Rosetta to calculate additional useful metrics by creating relax and interface analysis bash scripts, which we have shared. Additionally, we have developed a short python notebook for the analysis and visualization of AlphaFold2 (ColabFold) outputs, which we think will be useful for future iGEM teams looking to parse through and interpret the vast amount of data that is part of the AlphaFold output. This code is attached as a PDF on our wiki.

Wet Lab

Our wet lab team has developed the protocols and plasmids needed to test the affinity of modified ABMs for human serum albumin. While our results are not yet definitive, the protocols and components created by our team can serve as a foundation for future researchers working with human serum albumin.

Plasmid Design

We have designed plasmids for our protein that are codon-optimized for expression in E. coli. These sequences will serve as useful tools for researchers aiming to replicate and expand upon our findings.

Separation Technique

We have developed protocols for separating the human serum albumin-ABM complex from free-flowing albumins. This protocol will be beneficial for those seeking methods to separate proteins of similar size. For detailed instructions, please refer to the experiment page.

Linker Chemistry

We developed protocols for synthesizing a linker (maleimide-PEG-OH) that can connect fourth-generation cephalosporins with our modified ABM. Although we have yet to test the success of this synthesis, our work lays a foundational framework for future researchers interested in exploring fourth-generation cephalosporins. For detailed instructions, please refer to the experiment page.

Kinetic Modeling

Simbiology

Simbiology was used to create compartmental pharmacokinetic models and to visualize the results of these simulations. We built off of Simbiology to create a novel way of visualizing the results of Simbiology models by adding another dimension allowing us to elucidate the effects of unknown parameters on the pharmacokinetics of our drugs. Future teams can use this to determine how an unknown parameter will affect the pharmacokinetics of their drugs.

PK-Sim

PK-Sim was used to create physiologically based pharmacokinetic models. We used the fraction unbound parameter to simulate the effects of binding to albumin. Future teams can use this technique to model how changes to affinity to albumin or a1-acid glycoprotein affect a drug’s time in the body.