Devel

[huiwenke]

This PR introduces significant refactoring and improvements to the flexible alignment (flexalign) functionality, centralizing the core logic into a unified dispatch engine, introducing a powerful FATCAT-hybrid mode, and exposing fine-grained optimization controls for secondary structures.

Key Changes & Features:

• Unified Alignment Engine (flexalign_unified) & FlexAlignMode:
  Consolidated disparate flexible alignment workflows into a single entry routine named flexalign_unified. The execution flow is directly governed by the FlexAlignMode enum, which maps to the command-line -mm options as follows:

  • FLEX_STANDARD: The standard mode. Used by -mm 7 (passed with ss_opt = 0) and -mm 8 (passed with ss_opt = 1).
  • FLEX_BEST: Used by -mm 9. This mode dynamically overrides the single ss_opt to evaluate both secondary structure configurations and find the best result.
  • FLEX_FATCAT: Used by -mm 10. Dispatches execution to the new FATCAT-style flexible alignment pipeline (flexalign_fatcat_main).

• Secondary Structure Optimization (ss_opt) and -mm 7/8/9:
  Implemented an evaluation loop (for (int cur_ss_opt = start_ss; cur_ss_opt <= end_ss; cur_ss_opt++)) in flexalign_unified to control secondary structure-aware scoring.

  • -mm 7: Enables secondary structure optimization (ss_opt = 0).
  • -mm 8: Disables secondary structure optimization (ss_opt = 1).
  • -mm 9: Loops through both ss_opt = 0 (enabled) and ss_opt = 1 (disabled), benchmarking the alignments to automatically retain the structural configuration that yields the highest overall TM-score.

• FATCAT-style Flexible Alignment (flexalign_fatcat_main via -mm 10):
  Introduced a robust hybrid alignment module that synergizes FATCAT’s topological flexibility with US-align’s accurate TM-score optimization.

  • Algorithmic Workflow:
    1. Pre-evaluation: Executes a full-sequence baseline alignment (equivalent to -mm 9). If the structure already aligns excellently (global TM-score $\ge 0.85$), it early-exits to save computational resources.
    2. AFP Extraction & Merging: Precomputes intra-protein distance matrices, detects initial Aligned Fragment Pairs (AFPs) using Kabsch rotation, and merges them along diagonals.
    3. Dynamic Programming & Domain Splitting: Utilizes a dual dynamic programming approach with gap, twist, and RMSD penalty models (generate_bounds) to optimally chain AFPs and split domains at geometrically justified hinge points.
    4. US-align Block Refinement: Feeds the dynamically determined structural blocks back into the core US-align engine (execute_flexalign_with_fallback), optimizing the local sub-alignments while assembling them into the final global flexible alignment.

• CLI Enhancements:
  Exposed the -hinge option in the main USalign command-line help menu (previously commented out). Added internal parsing (hinge_set) so users can explicitly define the maximum number of hinges allowed during flexible alignments, or omit the -hinge option to remove the hinge limit for processing large and complex structures.