Back

Neretva: Unsupervised Neural Variational Inference for Allele-level Genotyping of Highly Polymorphic Genes

Zhou, Q.; Ahmadi, S. P.; Numanagic, I.

2026-02-05 bioinformatics
10.64898/2026.02.03.703582 bioRxiv
Show abstract

Accurate genotyping and phasing of highly polymorphic gene families are essential for precision medicine. Yet, the genotyping problem remains computationally challenging due to extreme sequence similarity between related genes, copy number variation, and structural complexity. Current methods typically rely on integer linear programming or maximum likelihood-based approaches that often suffer from scalability and flexibility issues. Here, we present Neretva, a new framework that models the genotyping problem as a probabilistic latent variable model and employs auto-encoding variational Bayes (AEVB) for inference. Leveraging neural inference networks, our approach scales efficiently to complex gene families such as CYP (cytochrome P450) and KIR (killer-cell immunoglobulin-like receptors), and achieves competitive or improved accuracy over the current state of the art on both families. Neretva is open source and is freely available at https://github.com/0xTCG/neretva.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.