Practical overview
When it fits
Use Diffabxl when evaluating or comparing computational methods for ranking antibody sequence designs, rather than when seeking a hosted design application.
Who it is for
- Researchers comparing generative models for antibody sequence design.
- Developers preparing a model interface for antibody-ranking benchmark evaluation.
Documented uses
- Compare log-likelihood ranking performance across antibody-design models and target datasets.
- Plan a Python interface that supplies antibody sequences and optional structure data to an evaluation pipeline.
- Review benchmark results across diffusion-based, language-model and graph-based approaches.
Requirements & limits
The implementation is available as a public GitHub repository under the Apache-2.0 license.
- The benchmark reports correlations across five target datasets, so performance can vary by dataset.
- The README describes its log-likelihood calculation as a naive approach used for consistency; calculations may differ by model type.
- The documented interface accepts sequences, optional PDB structure information, and model-specific parameters, and returns ranking scores with possible additional metrics.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Define a Python scoring interface for antibody-model benchmarking
Prepare a model interface that can be integrated with DiffAbXL’s evaluation pipeline.
Before you start
- A Python class and method for the model being integrated.
- A method able to accept a list of antibody sequences.
- If applicable, related structure data such as a PDB file, plus any other model-specific inputs.
Steps
Implement a method in a Python class that accepts antibody sequences as a list.
Add optional structure information and any additional model-specific parameters the model requires.
Return a dictionary containing log-likelihood scores and any other relevant metrics, such as RMSD, pAE, or model-specific outputs.
Expected outcome
A Python-method interface shaped for integration with the evaluation pipeline, returning sequence scores and other relevant metrics.
Guide limitations
- The repository recommends this interface but does not document execution or integration steps.
- Structure information is optional and model-dependent.
- The README notes that its log-likelihood scores use a naive approach for consistency; more principled approaches may vary by model type.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- AstraZeneca/DiffAbXL · Official homepage title · retained in its source language
Related options
DiffAb — The README identifies DiffAbXL as a re-implementation of the original DiffAb work.