# SNOOPPI: Sequence-Normalized Database of On- and Off-Target Protein-Protein Interactions

SNOOPPI is a curated protein–protein interaction (PPI) dataset containing positive, negative, and unknown interaction annotations. 
  * SNOOPPI is *sequence-indexed*, meaning one unique SNOOPPI PPI or non-PPI indicates a unique combination of protein sequences.
  * The dataset preserves protein sequences, identifiers, species information, PSI-MI interaction terms, experimental evidence, publications, and feature relationships where available.

## Dataset splits

SNOOPPI is distributed in three splits:

| Split      | Description                                                                                                                                 |
| ---------- | ------------------------------------------------------------------------------------------------------------------------------------------- |
| `positive` | Protein pairs retained as positive interactions under the SNOOPPI annotation criteria                                                       |
| `negative` | Protein pairs retained as negative interactions under the SNOOPPI annotation criteria                                                       |
| `unknown`  | Protein pairs present in IntAct's source interaction records but not assigned a definitive positive or negative label under the SNOOPPI criteria |

The `unknown` split should not automatically be interpreted as a negative or as a set of experimentally tested non-interactions. These entries represent interactions for which the available evidence was insufficient, conflicting, or otherwise unresolved under the annotation framework.

## Installation

Install the Hugging Face `datasets` package:

```bash
pip install -U datasets
```

## Load the complete dataset

```python
from datasets import load_dataset

snooppi = load_dataset("ChatterjeeLab/SNOOPPI")

print(snooppi)
```

This returns a `DatasetDict` containing the `positive`, `negative`, and `unknown` splits.

Access an individual split from the resulting object:

```python
snooppi_positive = snooppi["positive"]
snooppi_negative = snooppi["negative"]
snooppi_unknown = snooppi["unknown"]
```

## Load one split directly

Load only the positive interactions:

```python
from datasets import load_dataset

snooppi_positive = load_dataset(
    "ChatterjeeLab/SNOOPPI",
    split="positive",
)
```

Load only the negative interactions:

```python
snooppi_negative = load_dataset(
    "ChatterjeeLab/SNOOPPI",
    split="negative",
)
```

Load only the unknown interactions:

```python
snooppi_unknown = load_dataset(
    "ChatterjeeLab/SNOOPPI",
    split="unknown",
)
```

## Convert a split to pandas

```python
from datasets import load_dataset

snooppi_pos = load_dataset(
    "ChatterjeeLab/SNOOPPI",
    split="positive",
).to_pandas()
```

Converting a complete split to pandas loads that split into memory. Users working in memory-constrained environments may prefer to filter the Hugging Face `Dataset` before calling `.to_pandas()`.

## Example queries

### Retrieve direct human positive interactions

The PSI-MI term `MI:0407` denotes a direct interaction. The NCBI taxonomy identifier for humans is `9606`.

```python
from datasets import load_dataset

snooppi_pos = load_dataset(
    "ChatterjeeLab/SNOOPPI",
    split="positive",
).to_pandas()

direct_human_positive = snooppi_pos.loc[
    snooppi_pos[
        "unique_interaction_mi_terms"
    ].fillna("").str.contains(
        "MI:0407",
        regex=False,
    )
    & snooppi_pos[
        "partner_A_species"
    ].fillna("").str.contains(
        "9606",
        regex=False,
    )
    & snooppi_pos[
        "partner_B_species"
    ].fillna("").str.contains(
        "9606",
        regex=False,
    )
].copy()

print(
    f"Direct human positive interactions: "
    f"{len(direct_human_positive):,}"
)
```

### Find all positive interactions involving a UniProt accession

```python
from datasets import load_dataset

snooppi_pos = load_dataset(
    "ChatterjeeLab/SNOOPPI",
    split="positive",
).to_pandas()

uniprot_accession = "P00734"

uniprot_ppis = snooppi_pos.loc[
    snooppi_pos[
        "partner_A_uniprot_ids"
    ].fillna("").str.contains(
        uniprot_accession,
        regex=False,
    )
    | snooppi_pos[
        "partner_B_uniprot_ids"
    ].fillna("").str.contains(
        uniprot_accession,
        regex=False,
    )
].copy()

print(
    f"Positive interactions involving {uniprot_accession}: "
    f"{len(uniprot_ppis):,}"
)
```

## Notes on identifier searches

Identifier fields may contain multiple source identifiers or annotations within one string. The examples therefore use literal substring matching with `regex=False`.

For analyses requiring exact identifier membership, users should parse the relevant identifier field according to its delimiter structure before testing membership.

## Intended use

SNOOPPI may support:

* Evaluation of protein–protein interaction prediction methods
* Analysis of positive and negative interaction evidence
* Study of sequence and feature relationships between interacting proteins
* Retrieval of interactions by UniProt accession, species, or PSI-MI term

Users should preserve the distinction among positive, negative, and unknown annotations when constructing evaluation or training datasets.

## Citation

Citation information will be added upon publication of the accompanying manuscript.