ABTrans: A Transformer-based Model for Predicting Interaction between Anti-Aβ Antibodies and Peptides

Area of Science:

• Neuroscience
• Immunology
• Computational Biology

Background:

• Recent approval of anti-amyloid-beta (Aβ) antibodies for Alzheimer's disease highlights the need for understanding antibody-target interactions.
• Developing more effective and safer Alzheimer's treatments requires precise knowledge of antibody binding specificities.

Purpose of the Study:

• To develop a deep learning model, ABTrans, for predicting the binding interactions between anti-Aβ antibodies and peptides.
• To assess the cross-reactivity of approved anti-Aβ antibodies with other human amyloidogenic proteins.
• To identify potential off-target interactions for selected anti-Aβ antibodies.

Main Methods:

• Trained the ABTrans deep learning model on dodecapeptide sequences from phage display and known anti-Aβ antibody sequences.
• Classified antibody-dodecapeptide binding into four levels: not binding, weak, medium, and strong.
• Evaluated antibody cross-reactivity with other human amyloidogenic proteins and screened for off-target candidates.

Main Results:

• The ABTrans model achieved an accuracy of 0.83 in classifying binding abilities.
• Aducanumab and Donanemab showed the least cross-reactivity among the tested anti-Aβ antibodies.
• Systematic screening identified potential off-target candidates for eleven selected anti-Aβ antibodies.

Conclusions:

• ABTrans is a valuable tool for predicting anti-Aβ antibody-peptide interactions.
• Understanding and minimizing antibody cross-reactivity is crucial for developing safer Alzheimer's therapeutics.
• Further investigation into identified off-target candidates is warranted to ensure treatment safety.