AbDev

AbDev is a predictive modeling framework for monoclonal antibody (mAb) biophysical properties.
It integrates deep learning–derived spatial descriptors (DeepSP) with machine learning models to predict 12 critical developability-related properties directly from antibody variable region sequences.

---

Overview

AbDev combines:

• DeepSP – a deep learning-based model for generating spatial properties from antibody sequences
• A descriptor engineering pipeline
• Machine learning models trained to predict experimentally measured biophysical properties

This framework enables rapid in silico screening of antibody candidates prior to experimental validation.

---

Pipeline Workflow

1️⃣ Feature Preparation

Prepare a CSV file named:

Sequence_Info.csv

This file must contain the variable region sequences of the mAbs to be analyzed.

---

2️⃣ Generate Spatial Properties (DeepSP)

Run:

DeepSP.ipynb

DeepSP generates 30 spatial descriptors from antibody sequences.

Output:

DeepSP_descriptors_anarci2_Abdev.csv

---

3️⃣ Predict Biophysical Properties (AbDev)

Run:

AbDev.ipynb

Output:

Prediction_Result.csv

This file contains predictions for 12 biophysical properties, including developability-relevant metrics.

---

🔄 Update: Migration from ANARCI to ANARCII

AbDev has transitioned from ANARCI to ANARCII for antibody sequence numbering.

Install via:

pip install anarcii

Why this change?

• pip installable
• Improved compatibility with modern Python environments
• Simplified installation (no legacy HMMER dependency)
• Active maintenance

Important Note

Due to differences in numbering logic and backend implementation, minor variations in IMGT residue assignments may occur.

These changes may propagate to:

• Descriptor calculations
• Feature engineering steps
• Downstream prediction outputs

For strict reproducibility of earlier results, ensure the same numbering backend is used.

---

Environment Requirements

• TensorFlow == 2.12.0
• ANARCII

Example setup:

pip install tensorflow==2.12.0 anarcii

---

Reproducibility

To reproduce published results:

1. Use the specified TensorFlow version (2.12.0)
2. Ensure consistent antibody numbering backend
3. Regenerate spatial descriptors before prediction

---

Citation

If you use DeepSP:

Kalejaye, L., Wu, I.E., Terry, T., & Lai, P.K.
DeepSP: Deep Learning-Based Spatial Properties to Predict Monoclonal Antibody Stability
Computational and Structural Biotechnology Journal, 23:2220–2229, 2024.
https://www.csbj.org/article/S2001-0370(24)00173-9/fulltext

If you use AbDev:

Wu, I.E., Kalejaye, L., & Lai, P.K.
Machine Learning Models for Predicting Monoclonal Antibody Biophysical Properties from Molecular Dynamics Simulations and Deep Learning-Based Surface Descriptors
Molecular Pharmaceutics, 2024.
https://pubs.acs.org/doi/10.1021/acs.molpharmaceut.4c00804

---