[ENH] Add minimal AptaMCTS pipeline (MCTS-based aptamer recommendation)

[kris70lesgo]

Reference Issues/PRs

Fixes #21

---

What does this implement/fix? Explain your changes.

This PR adds a minimal implementation of AptaMCTS, an aptamer recommendation pipeline that combines a scoring model with the existing Monte Carlo Tree Search (MCTS) algorithm.

Key additions:

• AptaMCTSPipeline: inference-first pipeline with predict() and recommend() methods
• AptamerEvalAptaMCTS: experiment adapter bridging the pipeline with MCTS via evaluate()
• Minimal feature encoder in utils: placeholder iCTF-style encoding (pairs_to_features)
• Integration with existing MCTS without modifying its implementation

The pipeline follows the AptaTrans-style design:

• Uses MCTS through experiment.evaluate() only
• Does not require training inside the pipeline
• Assumes a pre-trained model exposing predict_proba()

---

What should a reviewer concentrate their feedback on?

• Correctness of MCTS integration via experiment adapter
• API consistency with AptaTrans (predict, recommend)
• Output format of recommend() (candidate, sequence, score)
• Placement and minimal design of preprocessing in utils
• Overall alignment with repository architecture (no MCTS modifications)

---

Did you add any tests for the change?

Yes.

Added focused tests covering:

• predict() returns correct type (np.float64) and uses predict_proba
• recommend() output structure and MCTS integration
• AptamerEvalAptaMCTS.evaluate() correctness
• Edge cases (invalid inputs, model validation)

All tests pass successfully.

---

Any other comments?

• The current feature encoding is a minimal placeholder, not a full iCTF implementation, as no reference implementation exists in the repository.
• This is intended to enable integration and can be extended in future work.
• MCTS implementation remains unchanged.

---

PR checklist

• The PR title starts with [ENH]
• Added/modified tests
• Used pre-commit hooks when committing to ensure that code is compliant with hooks

[NennoMP]

The issue specifically mentions to look at AptaNet interface (quoting myself I would expect a very similar interface to that of AptaNet) rather than AptaTrans. Indeed, I don't see methods such as fit or predict_proba. AptaTrans was mentioned purely in terms of compatibility w.r.t. MCTS.

[NennoMP]

Interface is not in line with AptaNet, please look carefully at the description.

[NennoMP]

Once the interface is similar to AptaNet as specified in the issue, this could easily be one single class that works for both AptaMCTS and AptaNet, since we expect a very similar interface.

[kris70lesgo]

@NennoMP Thanks for the review

• I’ve refactored the AptaMCTS pipeline to align with the AptaNet interface:
• Added fit, predict_proba, and predict methods
• Updated the experiment adapter to use predict_proba on (aptamer, target) pairs
• Ensured compatibility with MCTS via experiment.evaluate
• Updated tests accordingly

Would appreciate another review

[NennoMP]

I don't see a module in line with what _feature_classifier.py does for AptaNet, i.e. defining a classifier and regressor.