A Sample-Centric and Knowledge-Driven Computational Framework for Natural Products Drug Discovery

Modern natural products (NPs) research relies on untargeted liquid chromatography coupled with mass spectrometry metabolomics. Together with cutting-edge processing and computational annotation strategies, such approaches can yield extensive spectral and structural information. However, current processing workflows require feature-alignment steps based on retention time which hinders the comparison of samples originating from different batches or analyzed using different instrumental setups. In addition, there is currently no analytical framework available to efficiently match processed metabolomics data and associated metadata with external resources. To address these limitations, we present a new sample-centric and knowledge-driven framework allowing multi-modal data alignment - e.g. through chemical structures, biological activities, or spectral features - and demonstrate its value in exploring large and chemodiverse natural extract datasets. Here, the experimental data is processed at the sample level, matched with external identifiers when possible, semantically enriched, and integrated into a unified knowledge graph. The use of semantic web technology enables comparison of processed and standardized data, information, and knowledge at the repository scale. We demonstrate the utility of the developed framework, the Experimental Natural Products Knowledge Graph (ENPKG), to leverage the results obtained from screening 1,600 plant extracts against trypanosomatids and streamline the identification of new antiparasitic compounds. Thanks to its versatility, the proposed approach allows for a radically novel exploitation of metabolomics data. Semantic web technologies are a fundamental asset and we anticipate that their adoption will strongly complement the current computational metabolomics pipelines and enable the community to advance in the description of global chemodiversity and drug discovery projects.