Add DNABERT-2 promoter finetuning tutorial notebook

[Gonza10V]

Motivation

• Provide an end-to-end example showing how to finetune DNABERT-2 for binary promoter classification using Hugging Face Transformers and the Trainer API.
• Make it easy to reproduce and try locally by including dependency hints and a synthetic fallback dataset when a local CSV is not available.

Description

• Add a new tutorial notebook at notebooks/tutorials/06_dnabert2_promoter_finetuning.ipynb containing setup, data loading, preprocessing, tokenization, training, evaluation, and inference sections.
• Include a dependency-install cell and reproducibility seed plus model configuration targeting zhihan1996/DNABERT-2-117M with trust_remote_code=True for tokenizer and model loading.
• Implement robust data handling with required-column validation, sequence sanitization ([^ACGTN] -> N), a DatasetDict train/validation/test split, and a DataCollatorWithPadding-based data pipeline.
• Add compute_metrics (accuracy, F1, precision, recall), TrainingArguments tuned for finetuning, a Trainer instantiation, and example inference code for new sequences.

Testing

• Validated the notebook JSON structure with python -m json.tool notebooks/tutorials/06_dnabert2_promoter_finetuning.ipynb, which completed successfully.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d3b648021d

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]

Guard second split from invalid stratification

This split can raise a ValueError when temp_df has only one sample per class (which happens with the notebook’s built-in 6-row fallback dataset after the first 70/30 split), because train_test_split(..., stratify=...) requires at least 2 members per class. In that case the tutorial stops before training, so users without data/data_DNABERT/promoter_classification.csv cannot run the notebook end-to-end.

Useful? React with 👍 / 👎.