Antibody sequence design (AbMPNN)

AbMPNN is a deep learning-based antibody inverse folding model optimized for antibody sequence design. It leverages 3D backbone structural information to efficiently predict amino acid sequences that fold into the target conformation, with particular strength in handling the highly diverse CDRH3 region.

Compared to general-purpose protein design models, AbMPNN shows significant improvements in sequence recovery rate, structural robustness, and antibody framework recognizability. The model is fine-tuned on large-scale antibody structure data and sequence libraries, generating stable, designable candidate sequences with native antibody characteristics.

Upload a protein structure file in PDB or mmCIF format, and the system will automatically parse chain information. After selecting chains to design and setting parameters, the model will generate multiple optimized amino acid sequences, with scores (score = -log_prob, lower is better) and sequence recovery rates for each.

1. Upload Protein Structure File (PDB / mmCIF):

3. Design Parameters:

Model:
Num. Sequences:		Generate 1-10 sequences per target
Sampling Temp:		Higher temperature = greater sequence diversity
Backbone Noise:		Add Gaussian noise (A) to backbone coordinates for diversity
Random Seed:		0 = random each time; non-zero = fixed seed
Exclude AAs:		Checked amino acids will not appear in generated sequences

4. Constraints & Biases (JSONL, optional):

The JSONL fields below provide finer control than the UI controls above. If filled, they override the corresponding chain selection / AA exclusion settings above. Position numbering is 1-based and independent of residue numbering in the PDB file.

Chain design spec — overrides chain selection above (chain_id_jsonl):	Per-chain AA exclusion — overrides exclusion list above (omit_AA_jsonl):
Fixed positions (fixed_positions_jsonl):	AA composition bias (bias_AA_jsonl):
Per-residue position bias (bias_by_res_jsonl):	Symmetry/Tied positions (tied_positions_jsonl):

References

arXiv:2310.19513