ALCHIMIA: AI-Generated Molecules for CB2R and S1R Targets

A 1.7 MB dataset of molecules generated by the ALCHIMIA framework, a hybrid reinforcement learning and genetic algorithm system. The framework uses a vocabulary of 33 medicinal chemistry-inspired transformations to optimize synthetic accessibility, drug-likeness, and binding affinity for two pharmacological targets. Author Domenico Alberga published the dataset on figshare in April 2026.

Use Cases

• Benchmarking generative AI models for molecular design based on the described QED and SA scores.
• Training policy networks for reinforcement learning in chemistry based on sequences of molecular transformations.
• Optimizing multi-objective molecular properties like synthetic accessibility and drug-likeness using a genetic algorithm.
• Studying chemical lineages and diversity in population-based optimization for specific protein targets.

Strengths

• Dataset is derived from a framework using 33 defined medicinal chemistry transformations, providing interpretability.
• Molecules were generated for two specific, pharmacologically relevant targets: human Cannabinoid Receptor 2 and Sigma nonopioid intracellular Receptor 1.
• The associated ALCHIMIA code is freely available on GitHub, enabling reproducibility and extension.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale model training.
• Data may reflect methodological bias inherent to the specific ALCHIMIA framework and its chosen baselines.

Provenance

Source

figshare, author Domenico Alberga.

Collection Method

Generated by the ALCHIMIA computational framework combining reinforcement learning and a genetic algorithm.

Freshness

Last updated 2026-04-16 07:29:42; freshness should be verified.