antpack 0.2.5

A Python package for processing, manipulating and making inferences about antibody sequence data

AntPack

AntPack is a toolkit for antibody numbering, data processing, statistical inference and machine learning for antibody sequences. It is currently in active development -- more updates soon! For installation and how to use, see the docs.

What's new in version 0.2.5

Starting in version 0.2.0, there are only two dependencies, numpy and pybind. AntPack can now also indicate to you in its output which numbered positions are framework and CDR and can sort a list of position codes (which makes it easy to merge numbering for many different numbered sequences to generate a fixed length array or MSA file).

Antibody numbering

Numbering antibody sequences is an important precursor for many statistical inference / machine learning applications. AntPack is orders of magnitude faster for numbering antibody sequences than existing tools in the literature (e.g. ANARCI, AbRSA), while providing >= reliability. AntPack also provides tools for merging lists of position codes and for easy extraction of specific CDRs and framework regions.

V / J genes

Identifying the most similar human V / J gene sequences is useful for a variety of purposes. AntPack provides tools for determining which human V and J gene sequences are most similar to the variable region chain provided as input.

Humanness and developability

Minimizing the risk of immunogenicity is important for selecting clinical candidates. In AntPack v0.1.0, we introduce a simple, fully interpretable generative model for human heavy and light chains that outperforms all comparators in the literature on a large held-out test set for distinguishing human sequences from those of other species. This scoring tool can be used to score sequences for humanness, suggest modifications to make them more human, identify liabilities, and generate highly human sequences that contain selected motifs.

Finding developability liabilities

Some sequence motifs are known to be associated with developability issues -- certain motifs are known, for example, to be prone to N-glycosylation or deamidation. AntPack provides a tool for finding these "liability" motifs in an input sequence. Note that that identifying liabilities through finding motifs in this way is known to be prone to false positives (an N-glycosylation motif, for example, will not always be glycosylated). Still, these kinds of alerts can be useful for making yourself aware of potential developability issues.

Citing this work

If using AntPack in research intended for publication, please cite either the preprint:

Jonathan Parkinson and Wei Wang. 2024. For antibody sequence generative modeling, mixture models may be all you need. bioRxiv: https://doi.org/10.1101/2024.01.27.577555

or the final paper in Bioinformatics.