Our submission focuses on the generation of targeted antibodies that bind to Nipah virus glycoprotein G. We primarily utilized our new suite of antibody language models called peleke-1 (silico.bio/peleke), which generate heavy and light chain Fv sequences given an epitope-annotated antigen sequence as input.

We then co-folded the peleke-1-generated sequences and selected for candidates that were predicted to bind at the desired epitope (the binding site of ephrin-B2 from PDB 2VSM) with multiple molecular interactions with the antigen.

For some candidates, we utilized ProteinMPNN to generate additional derivative sequences based on a peleke-1-generated antibody or reference antibody (e.g. m102.3). The use of ProteinMPNN produced more varied sequences, some of which are predicted to exhibit desired binding properties.

Notes: All sequences are the heavy and light Fv chain sequences connected by a (G3S)4 linker. Also, all generated sequences were numbered through the Chothia numbering system to ensure that they all are antibody-like sequences.