Normalize asym_id to deterministic 0-based chain indices for multimer predictions

[DimaMolod]

Motivation

• Ensure chain indices used by protein.to_pdb are deterministic and 0-based because AlphaFold multimer asym_id values can be arbitrary positive integers and may change across runs.
• Avoid seed-dependent chain assignment differences that break reproducibility and PDB chain labeling for multimer predictions.

Description

• Add a new helper _normalize_asym_id that normalizes asym_id to contiguous 0-based indices using first-appearance order and handles both 1-based and 0-based encodings.
• Use a fallback_feature_dict when asym_id is missing from the processed features to recover original values.
• Replace the previous ad-hoc decrement (asym_id - 1) with calls to _normalize_asym_id in both the existing-results reconstruction path and the main prediction path.
• Improved docstrings and comments around template feature resetting for clarity.

Testing

• Ran the project's automated test suite with pytest -q, and the test run completed successfully.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 278097646c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Compute asym_id rank mapping from first appearance order

appearance_order = np.argsort(first_indices) produces a permutation from rank → unique-id index, but inverse_indices contains unique-id indices, so appearance_order[inverse_indices] uses the wrong direction unless that permutation happens to be self-inverse. For inputs where chain IDs appear in an order different from numeric sort (e.g. asym_id=[20,30,10,...]), this maps residues to the wrong normalized chain indices, so chain A/B/C labeling can be incorrect relative to sequence order and downstream chain-specific analyses can be misassigned.

Useful? React with 👍 / 👎.