EMBER-Embedding Multiple Molecular Fingerprints for Virtual Screening.

Area of Science:

• Computational chemistry
• Machine learning in drug discovery
• Bioinformatics

Background:

• Deep learning in virtual screening often debates neural embeddings versus classical descriptors.
• Graph Neural Networks are increasingly used to overcome limitations of short-range molecular fingerprints.
• Existing methods struggle to capture comprehensive molecular properties for accurate bioactivity prediction.

Purpose of the Study:

• To introduce EMBER, a novel molecular embedding technique.
• To evaluate EMBER's effectiveness in predicting ligand bioactivity using a deep convolutional architecture.
• To analyze the contribution of individual fingerprints within EMBER for target-specific predictions.

Main Methods:

• EMBER was developed using seven distinct molecular fingerprints arranged as "spectra".
• A deep convolutional neural network architecture was employed for bioactivity assessment.
• The model was trained and validated on a dataset of twenty protein kinases with binding sites similar to CDK1.

Main Results:

• EMBER demonstrated effectiveness in assessing ligand bioactivity against protein kinases.
• The study provides a detailed explanation of the deep convolutional architecture and its training.
• An explainability analysis was performed to understand the contribution of each fingerprint component.

Conclusions:

• EMBER offers a powerful new approach for molecular representation in deep learning-based virtual screening.
• The multi-spectral embedding strategy captures diverse molecular features crucial for predicting bioactivity.
• The findings support the use of EMBER for identifying potential drug candidates targeting specific protein kinases.