Preprint proposes method to extend protein variant effect predictions across untested genetic contexts

A preprint posted to bioRxiv presents a framework for extending the predictions of multiplexed assays of variant effects (MAVEs) — high-throughput experimental measurements of how amino acid substitutions alter protein function — to contexts not directly tested in the laboratory.

MAVEs can, in principle, measure the functional consequence of every possible single amino acid substitution in a protein of interest, but they are always conducted under specific conditions: a particular cell type, genetic background, or environmental setting. The authors note that the space of biologically relevant contexts — different tissues, disease states, co-occurring variants — is effectively infinite, making exhaustive experimental coverage impossible.

The preprint introduces a machine-learning approach that treats context as a structured variable and learns to generalise from sub-saturation contextual MAVE datasets. The method predicts how variant effect scores shift between contexts, enabling researchers to prioritise which additional experiments would most efficiently extend coverage.

This work is relevant to variant interpretation research, particularly efforts to understand why the same variant can have different functional consequences in different cellular environments — a known challenge for classifying variants of uncertain significance (VUS). The preprint has not yet undergone peer review.