SMILES Processing & Classification Utilities — Fork Updates

This fork provides major improvements to the SMILES-X utilities for molecular data processing and classification.
The updates focus on increasing robustness, preserving data integrity, and enhancing uncertainty estimation for more reliable model evaluation.

---

🧩 Overview

This modified version of the SMILES-X suite refines data preprocessing, feature encoding, and classification evaluation steps commonly used in molecular machine learning workflows.
It introduces better input handling, flexible sequence encoding, and uncertainty-aware classification metrics.

---

🚀 Key Changes & Enhancements

1. smiles_concat — Input Validation

• Enhanced validation to prevent misuse when a single SMILES string is passed instead of a list.
• Produces clear logging errors for incorrect input types.
• Ensures concatenation only occurs for lists or tuples of SMILES sequences using 'j'.

Example:

if isinstance(smiles, str): logging.error( "smiles_concat expected a list of SMILES per entry but got a STRING." ) logging.error("Wrap your SMILES into a list, e.g. ['CCO']")

text

✅ Result: Reduced runtime errors and improved robustness during SMILES batch processing.

---

2. int_vec_encode — Dynamic Padding (No Truncation)

• Replaced fixed-length truncation with dynamic padding to handle variable-length molecule sequences.
• Pads all SMILES to the maximum sequence length in the batch.
• Preserves critical information for longer SMILES strings.
• Supports 'unk' for unknown tokens and 'pad' for padding.

Example:

pad_len = max_length - len(ismiles) ismiles_tmp = ismiles + ['pad'] * pad_len

text

✅ Advantages:

• Prevents data loss from truncation.
• Compatible with transformer-based models (e.g., SMILES-BERT).
• Future-ready for attention mask integration.

---

3. sigma_classification_metrics — Monte Carlo Uncertainty Estimation

• Added Monte Carlo simulation for improved robustness in classification metrics.
• Injects Gaussian noise into predictions using predicted error estimates.
• Computes standard deviation across multiple stochastic runs for:
  • Accuracy
  • Precision
  • Recall
  • F1-score
  • PR-AUC

Example:

pred_mc = pred + np.random.normal(0, err_pred, size=len(pred)) metrics_mat[i] = [acc, prec, rec, f1, pr_auc]

text

✅ Impact:

• Produces uncertainty-aware metrics.
• Enables statistically meaningful interpretation of model performance.

---

4. Code Cleanup & Maintenance

• Pruned unnecessary imports from main.py.
• Improved readability, modularity, and general code hygiene.
• Reduced clutter, making it easier to extend or integrate new models.

---

⚙️ Example Usage

Safe concatenation of SMILES sequences smiles_concat(smiles_list)

Integer encoding with dynamic padding int_vec_encode(tokenized_smiles_list, vocab)

Monte Carlo uncertainty estimation for classification metrics sigma_classification_metrics(true, pred, err_pred, n_mc=1000)

text

---

📈 Impact Summary

Improvement	Benefit
Input validation in smiles_concat	Prevents misuse and clarifies error handling
Dynamic padding in int_vec_encode	Preserves complete molecule information
Monte Carlo uncertainty metrics	Enables uncertainty quantification in evaluation
Code cleanup and reorganization	Enhances maintainability and readability

---

🧠 Future Directions

• Integrate attention-mask generation for transformer compatibility.
• Extend uncertainty estimation to regression tasks.
• Add configuration utilities for dynamic hyperparameter control.

---

🧾 Notes

These updates were implemented as part of a research-oriented fork to improve reliability, interpretability, and extendability in molecular machine learning workflows.

---