A recent publication from researchers in Rice University’s Kavraki Lab demonstrates how a modeling tool has the potential to accelerate the development of immunotherapies for cancers and other diseases. The study, “APE-Gen2.0: Expanding Rapid Class I Peptide–Major Histocompatibility Complex Modeling to Post-Translational Modifications and Noncanonical Peptide Geometries” appears in the Journal of Chemical Information and Modeling.
The significance of the research, in a nutshell, is to “use computers to generate structures of proteins, and those structures we can use as information to guide the design of therapeutics,” says lead author Romanos Fasoulis. For this study, Fasoulis, a PhD student in Rice’s Department of Computer Science, expanded the existing APEGen tool so it could generate a greater variety of protein structures and peptides bound to protein structures, especially atypical peptides, which often show up in cancerous cells.
Peptides, or small strings of proteins, binding to the major histocompatibility complex (MHC) proteins is an important step in the immune response. When a peptide binds to the MHC, “the MHC acts as a transporter,” Fasoulis explains, and “it takes the whole peptide-MHC structure to the surface of the cell, so it’s exposed.”
Then a T-cell comes along, “scanning the surface of the [now exposed] peptide-MHC structure, and if it picks up something that it hasn’t seen before,” says Fasoulis, the T-cell determines whether the peptide is “self” or “nonself.” “If a cell is infected, or we have a tumor cell, some of these peptides will be different—these would be called ‘nonself,’ and the job of the T-cell is to identify this ‘nonself’ and elicit a cascade of responses, resulting in an immune response,” he says.
APE-Gen2.0 “allows us to screen and actually check for how some peptides can bind or not to the MHC Class 1 protein,” says co-author Maurício Menegatti Rigo, PhD. “This tool is useful for people who are interested in doing cancer research and immunotherapy.”
Corresponding author Lydia Kavraki, PhD—the Noah Harding Chair and professor of Computer Science, Electrical & Computer Engineering, Mechanical Engineering, and Bioengineering at Rice—explains the potential applications of APE-Gen2.0: “Personalized immunotherapies that leverage the adaptive immune system represent a rapidly evolving area of research with significant implications for cancer treatment. Our work develops computational tools that can be used independently or in combination, with the end goal of enabling researchers and clinicians to identify T-cells capable of targeting MHCs that display tumor peptides, thereby reducing the number of wet lab experiments needed to develop targeted therapies. This approach streamlines clinical pipelines, ultimately benefiting patients.”
The Evolution of APE-Gen2.0
In 2019, researchers at Rice introduced APE-Gen, a tool available on GitHub that, once a peptide sequence and MHC were input, generated “multiple clash-free conformations of a peptide bound to the MHC.” Its 2.0 iteration has improved “on three fronts”: modeling accuracy, an expanded peptide-MHC modeling repertoire, and third, Fasoulis adds, “the tool is more accessible than ever before.” It’s now provided as a web service, and, in only a few minutes, “with a few clicks and without any installations,” anyone can generate an accurate model of a pMHC structure.
“There were a bunch of cases APE-Gen could not model; the 2.0 iteration can model those cases, and those cases are of particular importance” says Fasoulis, referring to “noncanonical” cases of peptides with different geometric structures and peptides exhibiting post-translational modifications (PTMs), which alter the chemical composition, and, in turn, the peptide’s appearance.
Rigo uses the example of how wearing an earring “modifies” the appearance of the ear. “Our tool [APE-Gen2.0] allows us to check peptides without modifications or with modifications,” he says.
The paper’s authors note that noncanonical peptide properties “have been shown to be present in neoantigens, therefore, accurate structural modeling of these instances can be vital for cancer immunotherapy.”
Fasoulis says, “Part of the motivation for APE-Gen2.0 is to provide a computational platform where you can get a model of a peptide bound to an MHC, which is as accurate as it can be. Obtaining this information through experiments is expensive and time-consuming. Expanding the repertoire of what can be modeled can ultimately help clinicians in the design of therapeutic treatments."
Sequence and Structure
“Traditionally—because the experiments are easier to perform—we’ve had a lot of protein sequence data,” Fasoulis says. Think of a protein sequence as a string of letters denoting amino acids. But “there’s only so much you can do by looking at the linear amino acid sequence. By looking at the three-dimensional structure of the peptide-MHC complex, a lot of information is suddenly illuminated that you didn’t have access to before.” This additional information is key when making decisions about immunotherapy, for example.
The structural data that APE-Gen2.0 provides can also bridge knowledge gaps in sequence data. “We have millions of data points of peptide sequences; however, there are also instances of peptides that exhibit PTMs, and there’s not a lot of data for those. We show in the paper that we can use structure to accommodate this lack of data, and that using structure, we can produce better results in those cases where we don’t have lots of sequence data,” Fasoulis says. “When you have a lot of sequence data, sequence-based methods do really well. But you need structure to look at things with a magnifying lens.”
This isn’t to say that structure trumps sequence, though. “There’s interplay between the two,” says Fasoulis. “It’s important to be looking at both data modalities. Both are relevant in developing clinical tools.”
Kavraki agrees. “The landscape has shifted: even though sequence-based data are still abundant, structural data are becoming more and more accessible,” says Kavraki. “Rather than choosing between a sequence-based or structure-based method, it is crucial to integrate both perspectives. By simultaneously considering sequence and structure, we can significantly enhance our ability to gain new biological insights and develop innovative therapeutics. It is an exciting new era.”
Bridging the Wet Lab and the Computer Lab
In a similar way, computer modeling data does not diminish the importance of experimental data. Rigo describes the interplay as “a constant conversation” between the results of wet lab experiments and the output of computer programs. “We are uniting efforts. They are working together—the wet lab and the computational part,” he says.
It is crystal structures grown in the lab that provide the ground truth for the modeling tool. In turn, explains Rigo, the modeling tool can take 1,000 peptides identified in the lab and, because “there is not enough time or money to assess every protein” by X-ray crystallography, “filter out these 1,000 peptides to the top 20” with the highest binding affinity and immunogenicity scores.
“To determine if a peptide binds to an MHC or if a peptide will elicit an immune response, you need to do a lot of experiments, which are very costly and extremely lengthy,” Fasoulis says. “This is where the computational part comes in, to filter out those candidates that won’t work. We can focus on a smaller pool of candidates, accelerating the process of drug discovery.”
Speed and Accuracy
When Rigo was completing his PhD, he used a modeling tool that took three hours to model one MHC and one peptide. APE-Gen2.0 “can do that in minutes,” he says. “It’s a fast tool.
“This is very important, because when you consider clinical needs, we have thousands of peptides and MHCs that we have to evaluate, so having a tool that can actually be accurate and fast is very important.”
Fasoulis agrees that speed is one of the most beneficial features of APE-Gen2.0. “I think this is why APEGen is relevant,” he says. “It’s open source, and it’s rapid. Anyone can use it, and anyone can use it fast.”