Molecular Fingerprints for Robust and Efficient ML-Driven Molecular Generation
This work addresses the need for efficient and robust molecular generation in drug discovery, though it appears incremental by building on existing VAE approaches with new fingerprints and metrics.
The authors tackled the problem of generating drug-like molecules by proposing a molecular fingerprint-based variational autoencoder, which achieved a significant improvement in synthetic accessibility (ΔSAS = -0.83) and up to 5.9x computational efficiency compared to an existing SMILES-based method.
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 ($Δ\bar{SAS}$ = -0.83) and in computational efficiency up to 5.9x in comparison to an existing state-of-the-art SMILES-based architecture.