double_diff/README.md

# Double Diff Screening Pipeline

## Overview

This pipeline implements a dual-dimension differentiation screening strategy to identify prophage-encoded proteins that exhibit significant divergence from known Anti-CRISPR (Acr) proteins at both the sequence and structural levels. The objective is to discover highly confident novel Acr candidates that likely utilize unprecedented, independent mechanistic modalities.

## Screening Strategy

The pipeline employs an integrated sequence-structure filtering framework. By systematically evaluating and excluding proteins that demonstrate significant sequence homology or structural conservation with experimentally validated Acr proteins (2025-verified dataset), we isolate "double-divergent" candidates.

## Input Data

| Data Type | Source Description |
|-----------|-------------------|
| Query Sequences | Prophage-encoded proteins extracted from self-targeting bacterial genomes |
| Reference Sequence Database | 2025-verified experimentally validated Acr protein sequences |
| Reference Structure Database | 2025-verified experimentally validated Acr protein 3D structures |

## Screening Workflow

### Stage 1: Sequence Divergence Screening (BLASTP-based)

Candidate prophage proteins are subjected to homology screening against the 2025-verified Acr sequence reference database to isolate targets with independent evolutionary trajectories:

- **Alignment Tool**: BLASTP
- **Threshold Parameter**: E-value ≤ 1E-5
- **Evaluation Metrics**: Query Coverage (qcovs), Percent Identity (pident)
- **Exclusion Criteria**: qcovs ≥ 30 AND pident ≥ 30
- **Retention Criteria**: **No significant BLASTP hits** OR (qcovs < 30 OR pident < 30)

*Note: Proteins retained at this stage include those that yield absolutely no hits against the database, as well as those with weak alignments failing the exclusion criteria, ensuring true sequence novelty.*

**Associated Script**: `gain_rm_seq_similar_seqs.py`

### Stage 2: Structural Divergence Screening (Foldseek-based)

Sequence-divergent candidates from Stage 1 undergo three-dimensional structural comparison to evaluate fold novelty:

- **Alignment Tool**: Foldseek
- **Evaluation Metrics**:
  - `inhibition_type_probability`: Probability of structural fold similarity
  - `alntmscore`: Structural alignment TM-score
- **Screening Criteria**: `inhibition_type_probability` = **NA** AND `alntmscore` = **NA**
- **Length Filter**: Sequence length ≥ 50 amino acid residues (`seq_len` ≥ 50)

*Note: Foldseek does not output records for pairs without significant structural homology. Consequently, candidates lacking any structural match to the reference database are merged into the final dataset with `NA` (Not Available) for all structure-related metrics.*

### Stage 3: Result Integration and Output

Final screening results are compiled in `double_diff_foldseek.csv` with the following fields:

| Field | Description |
|-------|-------------|
| query | Candidate protein unique identifier |
| target | Best structural match reference protein (null if no valid match) |
| inhibition_type_probability | Structural fold similarity probability (0 indicates no structural match) |
| alntmscore | Structural alignment TM-score (0 indicates no significant alignment) |
| lddt | Local Distance Difference Test score |
| Acr_Type | Reference protein inhibition type classification (null if no match) |
| seq | Amino acid sequence |
| seq_len | Sequence length in amino acid residues (filtered for ≥ 50) |
| blastp_pident | BLASTP percent identity (0 for no hits/excluded) |
| blastp_qcovs | BLASTP query coverage (0 for no hits/excluded) |

## Screening Logic Diagram

```text
┌─────────────────────────────────────────────────────────────────┐
│              Prophage Protein Candidate Dataset                 │
└──────────────────────────┬──────────────────────────────────────┘
                           │
                           ▼
┌─────────────────────────────────────────────────────────────────┐
│  Stage 1: Sequence Homology Screening (BLASTP)                  │
│  - E-value ≤ 1E-5                                               │
│  - Exclude: qcovs ≥ 30 AND pident ≥ 30                          │
│  - Retain: No hits OR below exclusion thresholds                │
└──────────────────────────┬──────────────────────────────────────┘
                           │
              ┌────────────┴────────────┐
              ▼                         ▼
       ┌──────────────┐          ┌──────────────┐
       │  Sequence-   │          │  Sequence-   │  ← Exclude
       │  Divergent   │          │  Similar     │
       │  (Retain)    │          │  (Exclude)   │
       └──────┬───────┘          └──────────────┘
              │
              ▼
┌─────────────────────────────────────────────────────────────────┐
│  Stage 2: Structural Similarity Screening (Foldseek)            │
│  - alntmscore = NA (No Foldseek hit)                            │
│  - inhibition_type_probability = NA (No Foldseek hit)           │
│  - seq_len >= 50                                                │
└──────────────────────────┬──────────────────────────────────────┘
                           │
              ┌────────────┴────────────┐
              ▼                         ▼
       ┌──────────────┐          ┌──────────────┐
       │  Structure-  │          │  Structure-  │  ← Exclude
       │  Divergent   │          │  Similar     │
       │  ★ Final     │          │  (Exclude)   │
       │  Candidates  │          │              │
       └──────────────┘          └──────────────┘