1 of 13

Enhancing Predictability in Antibody-Antigen Interaction Characterization

Grace Niu, YSP Student - Bedford High School

Leo Murthy, YSP Student - Newton North High School

Natesan Mani, PhD Student - Northeastern University

Jason Kantorow, PhD Student - Northeastern University

Simran Pandey, Masters Student - Northeastern University

Prof. Srirupa Chakraborty, SimBioSys Lab - Northeastern University

1

2 of 13

​

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

Background

  • Antigens: foreign substances that trigger an immune response
  • Antibodies: proteins designed to protect the body from harmful entities by binding directly to antigens
  • Glycoproteins: Proteins that have glycans attached to their surface.
  • Glycans: Polysaccharides that are often located on the surface of proteins. They can alter the proteins structure, stability and ability to interact with other molecules
    • Glycans attached to the surface of antigens hinder antibodies from binding to antigens

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

2

Image Credit: https://www.britannica.com/science/antibody

Figure 1: Structure of an Antibody

Antibody

Spike Protein

Glycans

Virus

Figure 2: SARS CoV-2 spike protein

Image Credit: https://www.science.org/doi/full/10.1126/science.368.6491.564

3 of 13

​

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

Abstract

Motivation

  • Determining the structure of large, complex biomacromolecules by experimentation is costly and time-consuming
  • Ab-Ag binding is hindered by glycan shielding, which causes glycoproteins on the surfaces of viruses to block antibody binding
  • Google’s new AI-powered AlphaFold 3 can predict structures given a protein sequence

Objective

  • Use in silico modeling to provide higher accuracy models of glycoproteins
  • Evaluate how effectively AlphaFold 3 can predict protein structure and Ab-Ag interactions

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

3

Figure 4. 7L89 PDB

Glycan dynamics create a sort of “molecular armor” against antibody binding

Figure 3: Glycan Shielding

4 of 13

​

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

First Steps

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

  • Evaluated accuracy of AlphaFold 3: predicted structures of HIV and SARS viral proteins
  • Wrote code to automate CIF to PDB conversion

​

  • Methods and Tools
    • Navigating Linux terminal using commands
    • Using VI Editor
    • Using PyMOL, VMD, AlphaFold
    • Obtaining and interpreting an RMSD value

Low RMSD value of 0.655 indicates highly accurate AlphaFold 3 structural prediction

4

Figure 5: PyMOL Modeled protein

Figure 6: Code which converts all CIF files to PDB files

code for downloading pdbs

5 of 13

​

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

Challenges

Solutions

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

  • PyMOL was incorrectly aligning the experimental and prediction structures
    • Inputting all main strands and antibodies yields numerous possible outcomes

5

Figure 7:

Inaccurately aligned structures indicated by red lines

We lowered RMSDs of inaccurate AlphaFold predictions by inputting only one main strand and one antibody

Figure 8: Correctly aligned structures, pink structure from AlphaFold

→

6 of 13

​

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

6

Figure 9:

7 of 13

​

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

Moving Toward Scripting

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

Previously, we had been manually:

  • Downloading PDB files
  • Retrieving protein sequences and entering into AlphaFold
  • Aligning proteins modelled by AlphaFold to validated structures reported in literature

7

Figure 10: Code that retrieves sequences

8 of 13

​

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

8

Read PDB names in the Excel file

Create a directory in which to save PDB files

Rename file to match desired format

Iterate through all rows and columns of data to extract each chain, residue, etc. of the protein

YES :D

NO >:(

Download PDBS

Retrieve Sequences

Interpret data from Excel file in the standard format for Biopython

Retrieves PDB files from a folder

Save sequences to a new excel file

Is PDB file name in desired format?

Download PDB files

9 of 13

​

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

9

Figure 11: Spreadsheet output of code

Results

10 of 13

​

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

Our Contribution

  • Evaluated the accuracy of AlphaFold 3
  • Created a script effective for large datasets

Next Steps

  • AlphaFold 3 is not yet available for local installation
  • Running the program using AlphaFold 2 as an alternative is time-consuming and unreliable.

Outlook

  • Promising success of AlphaFold 3 with consideration of correct usage
    • Accurate modeling of protein structures can advance pharmaceutical development
  • Vaccine for HIV
    • Optimize results by designing a specific antibody strain to target a specific antigen

Conclusions

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

10

Image credit: https://www.europeanpharmaceuticalreview.com/news/141892/novel-hiv-vaccine-approach-shows-promise-in-landmark-first-in-human-trial/

Image Credit: https://www.digit.in/news/general/google-deepminds-new-alphafold-3-ai-can-model-proteins-dna-rna-details-here.html

11 of 13

Enhancing Predictability in Antibody-Antigen Interaction Characterization

NEU Young Scholars Program 2024 - July 31, 2024

Acknowledgements

SimBioSys Lab

Prof. Srirupa Chakraborty, Chemical Engineering, Northeastern University

Natesan Mani, Chemical Engineering, Northeastern University

Jason Kantorow, Chemical Engineering, Northeastern University

Simran Pandey, Chemical Engineering, Northeastern University

​

Center for STEM Education

Claire Duggan, Executive Director

Jennifer Love, Associate Director

Theodore Lourie, Angelina Le, Victoria Berry, Michael Marchev, YSP Coordinators

Nicolas Fuchs, Program Manager

Mary Howley, Administrative Officer

​

11

12 of 13

شكرًا لك

谢谢

धन्यवाद

நன்றி

감사합니다

ありがとう

متشکرم

Teşekkürler

Спасибо

תודה

Cảm ơn

אדאנק

ευχαριστώ

Thank you

Благодаря ти

ধন্যবাদ

13 of 13

Young Scholars Program at Northeastern University�Claire Duggan, Program Director

​

This work was supported by the National Science Foundation

Enhancing Predictability in Antibody-Antigen

Interaction Characterization

Grace Niu, YSP Student, Bedford High School​

Leo Murthy, YSP Student, Newton North High School​

Natesan Mani, Jason Kantorow, Simran Pandey, Department of Chemical Engineering, Northeastern University​

Prof. Srirupa Chakraborty, Department of Chemical Engineering, Northeastern University​

Abstract

Background

Experimental Methods

Conclusion and Future Steps

Results

Acknowledgements

  • Glycans are polysaccharides that can be located on the surfaces of proteins called glycoproteins. They can alter the protein’s structure and ability to interact with other molecules.
  • Glycans are frequently attached to the surface of viruses. This hinders the ability of an antibody to bind to an antigen and makes it difficult to combat viruses.
  • In silico modeling is crucial to Ab-Ag interactions because glycans are very dynamic, making it difficult to obtain high resolution experimental models of glycoproteins. Traditional methods such as X-ray crystallography and NMR spectroscopy are also costly and time-consuming.
  • In recent years, AI has presented itself as an efficient and versatile means of research. We are looking into AlphaFold, an AI-based program that is trained to predict protein structures given a sequence. Comparing the performance of AlphaFold to current methods for protein modeling is our main task.
  • Our research facilitates better understandings of Ab-Ag interactions, through AI-based computational modeling, allowing researchers to develop more effective drugs to treat viruses.

​

Aksimentiev, A., et al. (2019). “Using VMD.”

https://www.ks.uiuc.edu/Training/Tutorials/vmd/tutorial-

html/index.html.

Berman, H. M., Westbrook, J., Feng, Z., Gilliland, G.,

Bhat, T. N., Weissig, H., Shindyalov, I. N., & Bourne, P.

E. (2000). The Protein Data Bank. Nucleic acids research, 28(1), 235–242. https://doi.org/10.1093/nar/28.1.235.

Cohen, J., (2020). “The race is on for antibodies that stop the new coronavirus.”

Science Insider. https://www.science.org/doi/full/10.1126/science.368.6491.

564.

References

Simulation of Biomolecular Systems (SimBioSys) Lab

Srirupa Chakraborty, Assistant Professor

Natesan Mani, PhD Student

Jason Kantorow, PhD Student

Simran Pandey, Masters Student

​

Center for STEM Education​

Claire Duggan, Executive Director

Jennifer Love, Associate Director

Theodore Lourie, Angelina Le, Victoria Berry, Michael Marchev, YSP Coordinators​

Nicolas Fuchs, Program Manager

Mary Howley. Administrative Officer

Conclusion:

  • AlphaFold 3 is skilled at modeling Ab-Ag interactions if used correctly. This reveals AlphaFold’s ability to model proteins, despite limitations in modeling glycans.
  • The program performs best (with PyMOL) when given only simple structures, as PyMOL inaccurately aligned large structures

Broader Implications

  • Accurate modeling of proteins can advance pharmaceutical developments
  • Understanding the structure of proteins involved in diseases can lead to the design of more effective treatments, such as a vaccine for HIV
  • Closer examination of Ab-Ag interactions enables researchers to optimize results by designing a specific antibody strain to target a specific antigen

Next Steps

  • Our script can be used to create a program available to all
  • The program will include a script to align structures in pyMOL and calculate RMSD. This can be achieved after AlphaFold 3 becomes available for local installation. Running the program using AlphaFold 2 as an alternative is time-consuming and unreliable.

Work Flow Chart

Download PDB files

Extract protein sequences

Enter these sequences into AlphaFold 3

Interpret data and decide whether AlphaFold is effective

​

Align AlphaFold 3 structures to validated structures

​

Software

  • Linux Terminal
  • Ubuntu

Programs

  • PyMOL
  • AlphaFold3
  • Python
  • VI Editor
  • VMD

Motivation

  • Determining the structure of large, complex biomacromolecules by experimental methods such as X-ray crystallography and NMR spectroscopy is costly and time-consuming
  • Ab-Ag binding is hindered by glycan shielding, which causes glycoproteins on the surfaces of viruses to block antibody binding, thus posing challenges in treating these viruses
  • Google’s new AI-powered AlphaFold 3 can predict structures given a protein sequence, and it can handle more molecules beyond amino acids, including glycans

Inaccurately aligned

Correctly aligned

Initially, we entered the entire structure of each PDB file, including repeated sequences. However, this caused problems when aligning structures generated by AlphaFold 3 to validated structures reported in literature. We addressed this by simplifying our approach by only entering unique chains, which yielded lower RMSDs.

Result of script

In order to evaluate the effectiveness of AlphaFold 3, we aligned structures generated by AlphaFold to validated structures reported in literature. Root mean square deviation (RMSD) is a way to compare how similar two points are in a 3D environment. In the context of our project, we used RMSD to quantify how similar AlphaFold generated structures were to universally accepted structures. An RMSD of <2 indicates two highly similar structures.

​

Glycan 24,

Chain A

Antibody

Spike Protein

Glycans

Virus

Objective

  • Use in silico modeling to provide higher accuracy models of glycoproteins
  • Evaluate how effectively AlphaFold 3 can predict protein structure and Ab-Ag interactions

Results

  • AlphaFold 3 predicts highly accurate Ab-Ag structures. This could help researchers save time and resources.