Cdi Dti Dti Prediction Eval

---
name: cdi-dti-dti-prediction-eval
description: Evaluates a multi-modal deep learning framework's ability to predict drug-target binding interactions across standard, cross-domain, and cold-start scenarios. It probes the model's capacity to integrate textual, structural, and functional biological features for robust binary classification under distribution shifts and unseen entities. Use when the user wants to benchmark on BindingDB, Davis, or asks about evaluating this task. Reports AUROC.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2510.19520
  bibtex_key: li2025cdidti
  confidence: high
---

# cdi-dti-dti-prediction-eval

> CDI-DTI: A Strong Cross-domain Interpretable Drug-Target Interaction Prediction Framework Based on Multi-Strategy Fusion — Xiangyu Li et al. (arXiv:2510.19520, 2025)

## What this evaluates

Evaluates a multi-modal deep learning framework's ability to predict drug-target binding interactions across standard, cross-domain, and cold-start scenarios. It probes the model's capacity to integrate textual, structural, and functional biological features for robust binary classification under distribution shifts and unseen entities.

## Datasets

- **BindingDB** — total ?; splits: train (-1), test (-1)
- **Davis** — total ?; splits: train (-1), test (-1)

## Metrics

- `Accuracy` — range: [0, 1]
  - Ratio of correctly predicted binding interactions to the total number of predictions.
- `F1-score` — range: [0, 1]
  - Harmonic mean of precision and recall, balancing false positives and false negatives.
- `AUROC` **(primary)** — range: [0, 1]
  - Area under the receiver operating characteristic curve, measuring the model's ability to discriminate between binding and non-binding pairs across all classification thresholds.
- `AUPRC` — range: [0, 1]
  - Area under the precision-recall curve, emphasizing performance on the positive (binding) class, particularly useful for imbalanced datasets.

## Input / output format

**Input**: Drug-target pair represented by multi-modal features: textual embeddings (ChemBERTa/ProtBERT), structural graphs (SMILES/AlphaFold), and functional embeddings (MoT5/DeepGO/BioBERT).

**Output**: Binary classification label (binding vs. non-binding) or interaction probability.

## Scoring recipe

```python
def compute_metrics(y_true, y_pred_proba, threshold=0.5):
    y_pred = (y_pred_proba >= threshold).astype(int)
    acc = np.mean(y_true == y_pred)
    f1 = f1_score(y_true, y_pred, average='binary')
    auroc = roc_auc_score(y_true, y_pred_proba)
    auprc = average_precision_score(y_true, y_pred_proba)
    return {'Accuracy': acc, 'F1-score': f1, 'AUROC': auroc, 'AUPRC': auprc}
```

## Common pitfalls

- Dataset class imbalance requires careful handling of F1-score and AUPRC rather than relying solely on Accuracy.
- Cross-domain transfer (B→D, D→B) suffers from distribution shift, causing traditional single-modality models to degrade significantly.
- Cold-start scenarios (unDrug, unTarget, unPair) test generalization to completely unseen entities, which standard train/test splits do not capture.

## Evidence (verbatim from paper)

> We use accuracy, F1-score, AUROC (area under the receiver operating characteristic curve), and AUPRC (area under the precision-recall curve) as performance metrics.

## Citation

```bibtex
@misc{li2025cdidti,
  title={CDI-DTI: A Strong Cross-domain Interpretable Drug-Target Interaction Prediction Framework Based on Multi-Strategy Fusion},
  author={Xiangyu Li et al.},
  year={2025},
  note={arXiv:2510.19520}
}
```

- arXiv: 2510.19520