Natural product drug discovery requires the identification of known and novel compounds that display desirable bioactivities. Historically accomplished using bioactivity guided fractionation, followed by isolation and structure elucidation using NMR techniques, this workflow often resulted in significant investments that resulted in the identification of known and/or compounds not suitable for drug discovery and development, either due to poor drug-like properties, chemical tractability, or synthetic complexity. In order to overcome these issues, we have developed two novel tools: MS2Mol and MS2Prop to predict molecular structure and chemical properties solely from MS2 spectra with unparalleled accuracy and scalability for compounds outside of training sets. By combining the distinctive capabilities of mass spectrometry with machine learning these two groundbreaking algorithms overcome many of the traditional challenges of natural products drug discovery by allowing early evaluation and prioritization without the need for isolation and purification of the natural products from their native matrix.<br/>For MS2Mol we evaluated the utility of the chemical structure predictions through testing with medicinal chemists comparing the predicted structure with the ground truth actual structure with usefulness of the predicted structure as a prioritization tool as an outcome measure . The testing paradigm included random structures and predictions from both MS2Mol along with predictions from other programs that were blinded to the medicinal chemists. This testing revealed that MS2Mol produced useful predictions at a 72% rate and an avaerage processing time for a single prediction of 600 milliseconds which addresses scalability. Additionally, we can also use MS2 signatures to predict 10 physicochemical properties, including Lipinski, Veber, and other drug-like properties, such as log P, TPSA, and fraction sp3 carbons with high accuracy, using MS2Prop. The aggregated R-squared value across all 10 properties was 0.93 for known compounds and 0.74 for novel compounds. For the first time, these tools allow inquiry into the exceptionally diverse chemical space of natural products independent of time-consuming, and often unproductive, isolation. We will describe the technology in the context of examples from our work, compare and contrast our methods to other similar type approaches, and highlight specific use cases in lead generation.

MS2Mol and MS2Prop use tandem mass spectral (MS2) signatures and machine learning to enable <i>de novo</i> structure and chemical property predictions in plant natural product extracts

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.

MS2Mol and MS2Prop use tandem mass spectral (MS2) signatures and machine learning to enable de novo structure and chemical property predictions in plant natural product extracts