MolSafeEval: Benchmark for Detecting Risks in AI-Generated Molecules

The rise of generative artificial intelligence in drug discovery has accelerated the creation of new molecules with promising properties. However, most current benchmarks focus on structural novelty or property optimization, leaving aside a critical aspect: safety. MolSafeEval emerges as a pioneering benchmark specifically designed to evaluate the risks associated with AI-generated molecules, integrating toxicological databases, hazard rules, and reasoning based on large language models. This approach enables the detection of unsafe features, such as toxicity or reactivity, that could go unnoticed in conventional processes. Implementing robust verification systems is key for pharmaceutical and biotechnology companies to trust the results generated by their models. In this context, having technology providers that offer AI for businesses is essential to integrate these control mechanisms into workflows. Furthermore, safe molecule generation requires scalable and flexible infrastructure. AWS and Azure cloud services enable processing large volumes of data and running complex simulations, while custom software solutions facilitate the customization of evaluation pipelines. On the other hand, cybersecurity plays a relevant role in protecting the integrity of sensitive data used in these analyses. Likewise, incorporating Power BI and other business intelligence tools allows for clear visualization of risk patterns, supporting decision-making. At Q2BSTUDIO, we develop AI agents and custom applications that automate the detection of hazardous compounds, combining expert knowledge with the power of generative models. The combination of these technological capabilities not only drives safer molecular design but also establishes new standards of trust in AI-driven research.