How to Use Humatch

Humatch is a gene-specific antibody humanization method that operates on paired heavy and light chain variable domains rather than treating each chain in isolation. The paper emphasizes that preserving sensible VH/VL pairing while moving a lead toward human germline-like sequence space can improve the practical value of humanization for therapeutic antibodies.

On Neurosnap, researchers choose Mode for either Classification or Humanising and enter each antibody as heavy:light in Input Sequences. This makes the service useful both for quickly asking whether a candidate already looks human-like and for actively generating humanized sequence proposals when a non-human or weakly human lead needs redesign.

The workflow is best suited to early therapeutic optimization, where immunogenicity risk, expression, and sequence plausibility all matter before experimental back-mutation campaigns begin.

Humatch uses lightweight convolutional neural networks trained to evaluate heavy-chain humanness, light-chain humanness, and heavy-light pairing compatibility. That three-part design is important because a humanized chain pair should not only resemble human germline sequence patterns individually, but also remain compatible as a paired antibody variable region.

On Neurosnap, Classification mode is useful for screening diverse leads or repertoire-derived candidates, while Humanising mode produces redesigned heavy and light chains together with germline V-gene suggestions, edit counts, and chain or pair scores. Because the method follows IMGT-style alignment logic, complete VH and VL variable domains should be provided in the correct heavy:light order.

Researchers generally use Humatch as a prioritization and redesign layer. Strong candidates are the humanized pairs that improve humanness and pairing scores while minimizing unnecessary edits to residues likely to matter for binding or developability.

• Paired-chain modeling: Humatch humanizes heavy and light chains jointly instead of pretending each chain can be optimized independently.
• Classification and redesign modes: Mode lets researchers either screen leads for humanness or actively generate humanized candidates.
• Germline-aware output: V-gene suggestions, edit counts, and chain or pair scores make redesign decisions more transparent.
• Therapeutic workflow fit: The method is well matched to antibody lead optimization where reduced immunogenicity must be balanced against preserved function.