Abstract: We have developed a novel form of artificial intelligence (AI) useful for addressing a wide range of mechanistic questions in drug discovery including compound target ID/mechanism of action (MoA), target discovery/validation, mechanism of toxicity analyses, patient selection, drug repurposing, biomarker development, and more. This approach combines search engine algorithms with a massive knowledge graph containing diverse pharmacology and multi-omics data.
This approach brings together many important features, including:
- Transparency: The data supporting any mechanistic inference can be easily traced back to its original source, allowing scientists to fully leverage their expertise when interpreting results. - Concision: The most experimentally supported results are ranked first, regardless of how much data is searched or how many additional results are returned. - Scale: Search engine algorithms can be tailored to work seamlessly and rapidly over multiple large, heterogenous, and highly complex data sets, identifying instances where multiple data sources converge on the same answer. - Simplicity: The platform can be used via a familiar search bar interface, viewed as a node-edge graph, or accessed from a Large Language Model (LLM). - Robustness: Consensus results rise to the top while non-reproducible results fall to the bottom, making the system extremely robust to noise.
Analyses can be initiated with many different data types, including chemical structures, gene lists from multi-omics experiments, or lists of cell lines. Multiple approaches can be used together to identify mechanistic hypotheses that have experimental support from diverse sources.
This approach is unprecedented in its ability to generate actionable, experimentally supported mechanistic hypotheses based on massive amounts of highly complex biomedical and biochemical data. We will present case studies demonstrating real-world drug discovery applications and will discuss ways in which it complements and enhances machine learning based approaches, such as LLMs.
More information is available at www.plexresearch.com.