| # Antibody Liability Predictor | |
| This repository contains a PyTorch neural network for predicting antibody | |
| developability liabilities from ESM-2 embeddings. | |
| ## Model architecture | |
| - Input: 640-D (VH 320 + VL 320) | |
| - Output: 4 regression values | |
| - Architecture: MLP (128 → 64) | |
| ## Usage | |
| ```python | |
| import torch | |
| from model import LiabilityPredictor | |
| model = LiabilityPredictor(input_dim=640) | |
| model.load_state_dict(torch.load("liability_predictor.pt", map_location="cpu")) | |
| model.eval() | |
| # x should be shape (640,) or (batch, 640) | |
| y_pred = model(x) | |