Extend VizFold Inference and Visualization to Boltz

[smarru]

Context / Motivation

VizFold is designed to support interchangeable protein structure prediction models. Extending support to Boltz strengthens VizFold’s modular architecture and comparative analysis capabilities.

Tasks

Install and validate Boltz dependencies in the VizFold environment. Identify key inference stages and extract:

• Layer-wise activations
• Attention maps
• Structural outputs
• Adapt existing trace archive format to support Boltz
• Validate outputs against known examples

Expected Outcomes

• Working Boltz inference within VizFold
• Comparable inference traces aligned with OpenFold
• Documentation for running and analyzing Boltz models

Skills Involved

• ML model integration
• Reverse engineering research code
• Comparative model analysis