Antibody Developability Assessment

---
name: abdev
description: In-silico developability assessment for therapeutic antibodies and nanobodies. Three-layer pipeline: chemical liability scan, TAP profiling, Thera-SAbDab benchmarking. No GPU, no API keys. Works on CPU in ~60s.
version: 1.0.0
author: Junior Yu, Max Zhang
license: MIT
dependencies: [requests, pandas, numpy, matplotlib, abnumber]
metadata:
  hermes:
    tags: [antibody, developability, drug discovery, nanobody, protein engineering, TAP, PNAS 2019]
    repo: https://github.com/junior1p/AbDev
---

# AbDev: Antibody Developability Assessment

In-silico developability profiling for therapeutic antibodies and nanobodies.

## When to Use This Skill

- Screening antibody or nanobody sequences before wet-lab validation
- Identifying chemical liabilities (deamidation, oxidation, isomerization) early
- Benchmarking new antibody designs against approved clinical therapeutics
- AI agent-driven protein engineering workflows

## Quick Start

```python
from abdev_pipeline import run_abdev

# Single nanobody (VHH)
result = run_abdev(
    sequence_input="EVQLVESGGGLVQPGGSLRLSCAASGFNIKDTYIHWVRQAPGKGLEWVARIYPTNGYTRYADSVKGRFTISADTSKNTAYLQMNSLRAEDTAVYYCSRWGGDGFYAMDYWGQGTLVTVSS",
    name="my_nanobody",
    out_dir="abdev_results",
)

# Full IgG (VH:VL pair)
result = run_abdev(
    sequence_input="EVQLVESGGGLVQPGGSLRLSCAAS...:DIQMTQSPSSLSASVGDRVTIT...",
    name="my_igG",
    out_dir="abdev_results",
)
```

## CLI

```bash
# Single sequence
python abdev_pipeline.py --seq "EVQLV..." --name my_vhh

# VH:VL pair
python abdev_pipeline.py --vhvl "VH_sequence:VL_sequence" --name my_igG

# Batch from FASTA
python abdev_pipeline.py --fasta antibodies.fasta

# Skip Thera-SAbDab download (offline mode)
python abdev_pipeline.py --seq "..." --skip-benchmark
```

## Installation

```bash
pip install abdev
# or
git clone https://github.com/junior1p/AbDev.git
cd AbDev && pip install -r requirements.txt
```

## Three-Layer Assessment

### Layer 1: Chemical Liability Scanning

Six liability categories with IMGT numbering:

| Liability | Pattern | High-Risk Location |
|---|---|---|
| Deamidation | N[ST], Asn alone | CDR |
| Oxidation | Met, Trp, His | CDR |
| Isomerization | DG, DS motifs | CDR |
| Unpaired Cys | Odd cysteine count | Any |
| RGD Motif | Arg-Gly-Asp | Any |
| N-Glyc Site | Asn-X-Ser/Thr (X≠P) | CDR |

### Layer 2: TAP Profiling

Five metrics from Raybould et al. PNAS 2019, benchmarked against 242 clinical antibodies:

| Metric | Method | Amber Threshold |
|---|---|---|
| CDR Length | IMGT numbering | VH>17 or VL>17 |
| Hydrophobicity | Kyte-Doolittle GRAVY | z > 1.5 |
| Charge Asymmetry | \|net charge VH - VL\| | > 4.0 |
| Isoelectric Point | Theoretical pI | pI < 6.5 or > 9.0 |
| Proline Density | Framework prolines | > 8% |

### Layer 3: Thera-SAbDab Benchmarking

- k-mer ($k=3$) sequence similarity search
- ~30s per query against数千种临床抗体
- Flags: IP risk (similarity > 90%) or novelty risk (similarity < 40%)

## Output

```json
{
  "antibody_name": "Trastuzumab",
  "developability_score": 0.585,
  "warnings": [
    "Low germline identity may increase immunogenicity"
  ],
  "layer1": { "aggregation_risk": "LOW", "liability_count": 0 },
  "layer2": { "tap_z_scores": {...}, "amber_flags": ["Proline Density"] },
  "layer3": { "nearest_ab": "Trastuzumab", "similarity": 1.0 }
}
```

## Limitations

- Layer 1 is sequence-only, no structural context
- TAP thresholds from 2019 data, may not fit novel formats
- k-mer similarity cannot replace structural similarity analysis
- First run requires Thera-SAbDab download (~50MB)