Fgs Dti Prediction Eval

---
name: fgs-dti-prediction-eval
description: Evaluates a fine-grained selective similarity integration framework for drug-target interaction prediction. It tests the model's ability to dynamically weight multiple drug and target similarity views based on local interaction consistency to predict binary interaction labels. Use when the user wants to benchmark on Nuclear Receptors (NR), G-protein coupled receptors (GPCR), Ion Channel (IC), Enzyme (E), Luo, or asks about evaluating this task. Reports AUC.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2212.00543
  bibtex_key: liu2022fgs
  confidence: medium
---

# fgs-dti-prediction-eval

> Fine-Grained Selective Similarity Integration for Drug-Target Interaction Prediction — Liu et al. (2022) (arXiv:2212.00543, 2022)

## What this evaluates

Evaluates a fine-grained selective similarity integration framework for drug-target interaction prediction. It tests the model's ability to dynamically weight multiple drug and target similarity views based on local interaction consistency to predict binary interaction labels.

## Datasets

- **Nuclear Receptors (NR)** — total 175; splits: unspecified (-1); repo https://github.com/Nanfeizhilu/FGS_DTI_Prediction
- **G-protein coupled receptors (GPCR)** — total 1350; splits: unspecified (-1); repo https://github.com/Nanfeizhilu/FGS_DTI_Prediction
- **Ion Channel (IC)** — total 3201; splits: unspecified (-1); repo https://github.com/Nanfeizhilu/FGS_DTI_Prediction
- **Enzyme (E)** — total 4640; splits: unspecified (-1); repo https://github.com/Nanfeizhilu/FGS_DTI_Prediction
- **Luo** — total 1923; splits: unspecified (-1); repo https://github.com/Nanfeizhilu/FGS_DTI_Prediction

## Metrics

- `AUC` **(primary)** — range: [0, 1]
  - Area under the receiver operating characteristic curve, measuring the trade-off between true positive rate and false positive rate across classification thresholds.

## Input / output format

**Input**: Multiple similarity matrices for drugs (e.g., chemical structure, side effects) and targets (e.g., amino-acid sequence, PPI), combined with a binary interaction label matrix.

**Output**: Predicted interaction scores/labels for drug-target pairs, and integrated similarity weights per entity.

## Scoring recipe

```python
def compute_auc(y_true, y_pred):
    from sklearn.metrics import roc_auc_score
    return roc_auc_score(y_true, y_pred)
```

## Common pitfalls

- Dataset splits (train/val/test) are not specified in the provided text.
- Similarity matrices are updated with newer database versions (KEGG, DrugBank, ChEMBL), requiring careful version pinning for reproducibility.
- Evaluation metrics are not explicitly listed in the provided excerpt.

## Evidence (verbatim from paper)

> In the experiments, we utilize five benchmark DTI datasets, including four collected by Yamanishi [26], namely Nuclear Receptors (NR), G-protein coupled receptors (GPCR), Ion Channel (IC), and Enzyme (E), and one obtained from [22] (denoted as Luo).

## Citation

```bibtex
@misc{liu2022fgs,
  title={Fine-Grained Selective Similarity Integration for Drug-Target Interaction Prediction},
  author={Liu et al. (2022)},
  year={2022},
  note={arXiv:2212.00543}
}
```

- arXiv: 2212.00543