level: research
regnetagents is a multi-agent system that searches for regulatory drivers in cancer genomics. it works across two types of gene regulatory networks: bulk tumor networks from tcga and single-cell networks from the gremln project. for a given gene, the framework classifies it in both networks, filters for known cancer genes using oncokb, and assigns a mode of action for tumor-derived relationships. candidates are ranked by how consistently evidence appears across networks, with categories like both, tcga-only, or gremln-only.
the system is built as a langgraph dag workflow, meaning multiple ai agents coordinate to complete the analysis. it does not infer new networks but acts as a downstream analytical layer over precomputed networks. users can access it through a python api or a model context protocol client, making it easy to integrate into existing bioinformatics pipelines. the framework is designed for structured, query-driven searches, so researchers can focus on specific genes of interest.
by combining bulk and single-cell data, regnetagents helps pinpoint regulatory relationships that might be missed when using only one data type. the mode-of-action assignment adds functional context, showing whether a regulator activates or represses its target. this can guide experimental validation and drug target discovery. the multi-agent design allows the system to handle complex, multi-step reasoning without manual intervention, speeding up the analysis of large cancer genomics datasets.
why it matters: it automates the search for cancer gene regulators across different data types, saving time and surfacing candidates that might otherwise be overlooked.