AI in Scientific Discovery Summit
- 4 days ago
- 1 min read
Last week, I had the pleasure of participating in the AI Scientific Discovery Summit at The Engine, organized by SandboxAQ.
It sometimes feels as though we are already living in the future. Large language models can now initiate and orchestrate scientific workflows using MCP server, run predictive models, and return results to scientists within minutes or even seconds for some applications. However, keeping a human in the loop is still essential. Scientists must verify simulation results, spot-check model outputs, and identify incorrect assumptions or parameters that could lead to unreliable conclusions.
At the summit, I presented our poster on AI and physics-based methods for target identification and discovery.
One area receiving particular attention in biotechnology is biomedical knowledge graphs. Biomedical data is highly heterogeneous and often unstructured. By integrating information from multiple sources while preserving the data provenance, it is possible to construct large, interconnected graphs spanning molecular mechanisms, genes, proteins, diseases, and even patient-level data.
These graphs can help reveal connections that may not be immediately apparent from individual publications or databases. They can also provide a foundation for more grounded scientific reasoning, allowing LLM-based agents to work with traceable evidence and helping scientists to identify and prioritize more promising targets and biomarkers.
Overall, the future of AI-driven drug discovery is coming. When large-scale virtual screening and physics-based modeling are combined with rigorous target selection, it could improve the probability of success in later stages of drug development.
There is still a lot of work to do, especially around experimental validation and scientific reliability, but I am excited to contribute to this work at SandboxAQ.
More information about the summit:
Comments