SyntheMol: Generative AI Data and Models for Novel Antibiotic Design

Kyle Swanson and colleagues from Stanford University published data and models in 2024 for generative AI-driven antibiotic discovery. The repository includes training data for Acinetobacter baumannii inhibition, known antibiotic molecules from ChEMBL, and compounds generated by SyntheMol using models like Chemprop and random forests. It also contains information on 58 SyntheMol-generated compounds that were successfully synthesized by Enamine.

Use Cases

• Training antibiotic property prediction models based on Acinetobacter baumannii inhibition data.
• Evaluating the novelty of generated molecular candidates based on comparison to known antibiotics from ChEMBL.
• Generating new antibiotic candidates using provided Chemprop, Chemprop-RDKit, and random forest models.
• Analyzing the synthesizability and properties of AI-generated compounds based on data from the Enamine REAL Space.

Strengths

• Includes data on 58 successfully synthesized compounds generated by the AI model.
• Provides multiple model ensembles (each an ensemble of ten models) for property prediction and generation.
• References specific, versioned external data sources like the Enamine REAL Space (2022 q1-2).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Stanford University, associated with the 2024 Nature Machine Intelligence paper by Swanson et al.

Collection Method

Data includes experimental inhibition assays, known molecules from ChEMBL, and AI-generated compounds via the SyntheMol framework.

Time Range

The Enamine REAL Space details are from version 2022 q1-2, downloaded on August 30, 2022.