StruSel

Structurome-wide Selectivity of pathogen-exclusive targets

---

Overview

StruSel is a computational pipeline for identifying pathogen-exclusive drug target candidates through proteome-scale structural comparison. It goes beyond traditional sequence-based homology exclusion by leveraging AI-predicted 3D structures to uncover proteins that are structurally divergent from the human proteome — representing a largely unexplored space for selective antimicrobial drug discovery.

Given a target pathogen, StruSel:

1. Downloads AlphaFold2-predicted structures for the pathogen proteome
2. Applies pLDDT-based quality filtering (≥ 70) to retain high-confidence structures
3. Performs structurome-wide structural similarity search against the human structurome using FoldSeek (3Di + GPU)
4. Performs parallel sequence-based search using MMseqs2 for benchmarking
5. Computes ESM2 protein language model embeddings for embedding-space similarity (prototype)
6. Classifies all proteins into a seven-category scheme (A/B/B2/C/C2/D/D*) based on sequence–structure concordance
7. Cross-references with essential gene databases (DEG) to filter for essential proteins
8. Maps candidates to CARD (via RGI) to assess AMR relevance
9. Outputs a tiered, annotated list of pathogen-exclusive target candidates

---

Results: Klebsiella pneumoniae vs Human Structurome

Input Scale

Proteome	Total structures	pLDDT ≥ 70	Source
K. pneumoniae (UP000007841)	5,727	5,358	AlphaFold DB v6
H. sapiens (UP000005640)	23,586	15,716	AlphaFold DB v6
H. sapiens (sequence)	205,155	—	UniProt

Seven-Category Classification

Category	Description	Count
A	Seq-homolog + Struct-homolog	445
B	Seq-similar + Struct-divergent	13
B2	Seq-similar + Struct-partial	5
C	Structural mimic (no seq hit)	1,167
C2	Partial structural mimic (no seq hit)	287
D	StruSel candidate (no seq hit, low TM)	151
D*	High-confidence StruSel candidate (no seq/struct hit)	3,290
	Total	5,358

Target Tiers

Tier	Definition	Count
Tier 1	Essential + D/D* + no AMR (prime targets)	1,412
Tier 2	Essential + D/D* + AMR-associated	16
Tier 3	Essential + C/C2 + no AMR (structural mimics)	895
Tier 4	Essential + C/C2 + AMR-associated	7

Tier 1 Functional Profile

• Enzymatic targets: 277 / 1,412 (19.6%)
• Membrane-associated: 466
• Transcription regulators: 219
• Transporters: 355
• Top molecular functions: DNA-binding transcription factor activity (171), ATP binding (155), metal ion binding (113)

Pipeline Parameters

Parameter	Value
pLDDT cutoff	≥ 70.0
MMseqs2 e-value	1e-5
MMseqs2 sensitivity	7.5
Query/target coverage	≥ 0.7
FoldSeek e-value	0.001
TM-score homolog	≥ 0.5
TM-score unrelated	< 0.3
ESM2 cosine similarity	≤ 0.5

---

Use Case

Current implementation: Klebsiella pneumoniae (WHO critical priority pathogen)

Planned extension: Pan-ESKAPE (E. faecium, S. aureus, K. pneumoniae, A. baumannii, P. aeruginosa, Enterobacter spp.)

---

Pipeline Overview

Pathogen proteome (AlphaFold DB)
        │
        ▼
pLDDT quality filter (≥ 70)
        │
        ├──────────────────────────────────┐
        ▼                                  ▼
MMseqs2 sequence search          FoldSeek structural search (GPU)
        │                                  │
        ├──────────────────────────────────┤
        ▼                                  ▼
  Sequence hits/non-homologs     Structural hits (TM-score bins)
        │                                  │
        └──────────┬───────────────────────┘
                   ▼
    Seven-category classification (A/B/B2/C/C2/D/D*)
                   │
                   ▼
    Essential gene filtering (DEG)
                   │
                   ▼
    AMR mapping (CARD-RGI)
                   │
                   ▼
    Tiered target candidates (Tier 1–4)
                   │
                   ▼
    UniProt GO/pathway annotation

---

Repository Structure

strusel/
├── README.md
├── LICENSE
├── notebooks/
│   └── strusel_pipeline.ipynb    # Full Colab pipeline (12 cells)
├── scripts/
│   └── .gitkeep
└── results/
    ├── strusel_classification.tsv
    ├── strusel_tier1_prime_targets.tsv
    ├── strusel_tier2_amr_targets.tsv
    ├── strusel_tier3_mimic_risk.tsv
    ├── strusel_tier4_amr_mimic.tsv
    ├── strusel_top_amr_targets.tsv
    ├── tier1_uniprot_annotations.tsv
    └── StruSel_Report_Klebsiella.pdf

---

Quick Start (Google Colab)

The pipeline runs end-to-end on Google Colab (GPU runtime, ~45 min). All dependencies are installed within the notebook.

---

Dependencies

• Python 3.8+
• FoldSeek (GPU-enabled)
• MMseqs2
• DIAMOND (for RGI)
• ESM2 (fair-esm)
• RGI (CARD)
• Biopython, pandas, numpy, matplotlib, seaborn

---

Citation

If you use StruSel, please cite:

Pranavathiyani G. StruSel: Structurome-wide Selectivity of pathogen-exclusive targets. GitHub. https://github.com/pranavathiyani/strusel

---

Author

Pranavathiyani Gnanasekar Assistant Professor (Research), Division of Bioinformatics SASTRA Deemed University, Thanjavur pranavathiyani@scbt.sastra.edu Google Scholar | ORCID

---

License

MIT License. See LICENSE for details.