[Paper Review] ADCNet: a unified framework for predicting the activity of antibody-drug conjugates

Liye Chen, Biaoshun Li+6 more|arXiv (Cornell University)|Jan 17, 2024

HER2/EGFR in Cancer Research5 citations

TL;DR

ADCNet is a unified deep learning framework that combines protein-language models (ESM-2) and small-molecule models (FG-BERT) to predict ADC activity from protein sequences, linker/payload SMILES, and DAR values, achieving strong test performance.

ABSTRACT

Antibody-drug conjugate (ADC) has revolutionized the field of cancer treatment in the era of precision medicine due to their ability to precisely target cancer cells and release highly effective drug. Nevertheless, the realization of rational design of ADC is very difficult because the relationship between their structures and activities is difficult to understand. In the present study, we introduce a unified deep learning framework called ADCNet to help design potential ADCs. The ADCNet highly integrates the protein representation learning language model ESM-2 and small-molecule representation learning language model FG-BERT models to achieve activity prediction through learning meaningful features from antigen and antibody protein sequences of ADC, SMILES strings of linker and payload, and drug-antibody ratio (DAR) value. Based on a carefully designed and manually tailored ADC data set, extensive evaluation results reveal that ADCNet performs best on the test set compared to baseline machine learning models across all evaluation metrics. For example, it achieves an average prediction accuracy of 87.12%, a balanced accuracy of 0.8689, and an area under receiver operating characteristic curve of 0.9293 on the test set. In addition, cross-validation, ablation experiments, and external independent testing results further prove the stability, advancement, and robustness of the ADCNet architecture. For the convenience of the community, we develop the first online platform (https://ADCNet.idruglab.cn) for the prediction of ADCs activity based on the optimal ADCNet model, and the source code is publicly available at https://github.com/idrugLab/ADCNet.

Motivation & Objective

• Motivate rational design of antibody-drug conjugates (ADCs) by better linking structure to activity.
• Develop a unified model that integrates protein sequences and small-molecule representations to predict ADC activity.
• Provide an online platform and open-source code to facilitate community access and reproducibility.

Proposed method

• Integrates protein representation learning with ESM-2 and small-molecule representation learning with FG-BERT.
• Incorporates antigen/antibody protein sequences, SMILES strings of linker and payload, and DR value (DAR) as inputs.
• Trains and evaluates on a carefully designed ADC dataset with ablation and external testing.
• Measures performance using accuracy, balanced accuracy, and AUC on the test set.
• Compares ADCNet against baseline machine learning models across multiple metrics.

Experimental results

Research questions

• RQ1 Can a unified framework combining protein and small-molecule embeddings improve ADC activity prediction beyond baseline models?
• RQ2 How well does the model generalize across different test settings and external data?
• RQ3 What is the contribution of each input modality (proteins, SMILES, DAR) to predictive performance?

Key findings

• ADCNet achieves an average test accuracy of 87.12%.
• ADCNet achieves a balanced accuracy of 0.8689 on the test set.
• ADCNet achieves an area under the ROC curve of 0.9293 on the test set.
• Cross-validation, ablation studies, and external testing support robustness and stability of ADCNet.
• An online platform is provided for ADC activity prediction using the optimal ADCNet model; source code is publicly available.