# Guiding AlphaFold predictions with experimental knowledge to inform dynamics and interactions with VAIRO

## TL;DR

This paper introduces VAIRO, a method to enhance AlphaFold predictions by incorporating experimental knowledge to better capture protein dynamics and interactions.

## Contribution

VAIRO is a novel method that integrates site-specific variants, alignments, and templates to guide AlphaFold predictions with experimental data.

## Key findings

- VAIRO rescues asymmetric and weaker interactions in bacterial surface layer assemblies.
- The method reveals dynamic states in pneumococcal membrane protein complexes previously inaccessible.
- VAIRO is publicly available via PyPI and GitHub for broader use.

## Abstract

Structural predictions have reached unprecedented accuracy. They leverage sequence‐specific data to capture all potential interactions a sequence has evolved to fulfill. AlphaFold derives information from three sources: learned parameters capturing intrinsic amino acid secondary structure and environment propensity; models of related proteins providing structural templates; and aligned sequences encoding profiles and concerted evolutionary changes of residues involved in contacts. However, function demands dynamic changes; hence not all possible interactions can coexist simultaneously. Comprehensive information entails contradictions, which resolved in favor of the better‐informed structure will silence less stable states and associations. Here, we introduce a method using all three channels to include prior knowledge: site‐specific variants, predefined alignments and templates. Selecting information relevant to a particular state delimits the functional context of a prediction. Our program VAIRO allows us to rescue asymmetric and weaker interactions to complete the view of molecular assemblies in the architecture of a bacterial surface layer, and reveals otherwise inaccessible dynamic states in a pneumococcal multimeric membrane protein complex. VAIRO is distributed via the python package index (PyPI) (https://pypi.org/project/vairo) and the code is also available on Github (https://github.com/arcimboldo-team/vairo).