Machine learning framework to extract physicochemical features of B-cell epitopes recognized by a cross-reactive...

Area of Science:

• Immunology and Parasitology
• Computational Biology and Bioinformatics

Background:

• Plasmodium falciparum infection during pregnancy involves the virulence factor VAR2CSA, mediating infected red blood cell adhesion to the placenta.
• Developing vaccines against malaria in pregnant women requires antibodies targeting VAR2CSA, but its polymorphism necessitates identifying conserved epitopes for broad immunity.
• A cross-reactive mouse antibody (3D10) against Duffy binding protein region II (DBPII) unexpectedly recognized diverse VAR2CSA alleles, suggesting shared epitopes within Duffy binding-like (DBL) proteins.

Purpose of the Study:

• To develop a computational framework for analyzing complex peptide array data to identify conserved epitopes.
• To investigate shared epitopes across different Plasmodium Duffy binding-like (DBL) proteins recognized by a cross-reactive antibody (3D10).
• To explore the potential of machine learning in epitope discovery for vaccine development against polymorphic parasite antigens.

Main Methods:

• Screened peptide arrays of four DBL proteins (including VAR2CSA alleles and DBPII) with the cross-reactive antibody 3D10.
• Developed and applied a machine learning framework utilizing decision trees to extract features correlated with 3D10 binding from array data.
• Validated model predictions using independent datasets (rodent Plasmodium DBL protein) and structural mapping, including testing mutant peptides and recombinant antigens.

Main Results:

• The machine learning framework successfully analyzed complex peptide array data, identifying features associated with 3D10 antibody binding.
• The approach predicted both linear and conformational epitopes, with conformational epitopes validated using recombinant proteins (e.g., PcDBP).
• Physicochemical properties of epitopes recognized by cross-reactive antibodies were successfully extracted from peptide array data.

Conclusions:

• Machine learning provides a powerful tool for mining peptide array data to identify conserved epitopes across polymorphic parasite proteins like VAR2CSA.
• The study identified shared epitopes within the Duffy binding-like protein family, crucial for developing strain-transcending malaria vaccines for pregnant women.
• This computational approach enables the discovery of conformational epitopes, enhancing our understanding of antibody recognition and vaccine design strategies.