Molecular Fingerprints for Robust and Efficient ML-Driven Molecular Generation

TL;DR

This paper introduces a new molecular fingerprint-based variational autoencoder that improves the diversity, drug-likeness, and synthetic accessibility of generated molecules while significantly enhancing computational efficiency over existing models.

Contribution

The authors develop a novel molecular fingerprint-based autoencoder with pharma-relevant metrics, achieving better synthetic accessibility and efficiency than prior SMILES-based approaches.

Findings

Improved chemical synthetic accessibility ($ar{SAS}$ decrease of 0.83)

Up to 5.9x faster computation

Enhanced diversity and drug-likeness of generated molecules

Abstract

We propose a novel molecular fingerprint-based variational autoencoder applied for molecular generation on real-world drug molecules. We define more suitable and pharma-relevant baseline metrics and tests, focusing on the generation of diverse, drug-like, novel small molecules and scaffolds. When we apply these molecular generation metrics to our novel model, we observe a substantial improvement in chemical synthetic accessibility ($\Delta\bar{{SAS}}$ = -0.83) and in computational efficiency up to 5.9x in comparison to an existing state-of-the-art SMILES-based architecture.