Add US-align structural comparison and evaluation module

Description of feature

Summary

This issue proposes adding US-align as a new structural evaluation module
This work will be started during the nf-core Hackathon March 2026.

Background

Currently nf-core/proteinfold predicts protein structures but provides
limited built-in tools to evaluate prediction quality against known
reference structures.

US-align fills this gap by providing:

• RMSD (Root Mean Square Deviation) — atom-level structural difference
• TM-score — global structural similarity (0 to 1, higher = more similar)
• Full structural alignment between predicted and reference structures

This is especially valuable when a reference experimental structure exists
(e.g. from the PDB) and researchers want to quantify how accurate a prediction is.

Proposed implementation

• No official Docker image exists — custom Dockerfile required (compile from C++ source)
• Source: https://github.com/pylelab/USalign
• Input: two PDB files (predicted + reference)
• Output: RMSD score, TM-score, alignment file
• New nf-core module: usalign/align

Resources

• US-align source: https://github.com/pylelab/USalign
• US-align paper: https://www.nature.com/articles/s41592-022-01585-1

Tasks

• Write custom Dockerfile for US-align (compile from source)
• Push image to GitHub Container Registry or Docker Hub
• Write nf-core module for usalign/align
• Connect as optional evaluation step after structure prediction
• Test with predicted vs reference PDB pair
• Document usage in pipeline README

Start with the US-align source repository and the proposed custom Dockerfile, then review how the usalign/align nf-core module should accept predicted and reference PDB files. Done means the image is published, the module produces RMSD, TM-score, and an alignment file for a PDB pair, the optional evaluation step is connected, and usage is documented in the pipeline README.