Ai-driven pesticide discovery using differential mobility spectrometry and machine learning

Overview

WaterMine Innovation offers a computational platform that integrates differential mobility spectrometry (DMS) data with ensemble stacking machine learning to predict the physicochemical properties of pesticide molecules and classify structurally related compounds as potential pesticides. By probing how molecules behave under microsolvation conditions in the DMS instrument, the system captures rich structural and interaction information that is then fed into proprietary molecular representations and combined ML models. This approach delivers high-accuracy property prediction and supports AI-guided structural optimization to remove undesirable characteristics such as toxicity.

The platform serves agrochemical and crop protection organizations seeking faster, more reliable ways to discover and refine pesticide candidates. It can reduce reliance on costly wet-lab screening by prioritizing molecules with the highest likelihood of success before synthesis and field testing.

Technical specifications

• DMS-ML workflow: Differential mobility spectrometry captures microsolvation behavior of molecules, providing structural insight beyond conventional analytical techniques.
• Ensemble stacking: Multiple ML classification and regression models are combined through a meta-analysis step that balances bias and variance to minimize prediction error.
• Proprietary molecular representations: WaterMine-developed descriptors complement standard chemical fingerprints, improving model generalization across molecular classes.
• Property prediction: Predicts physicochemical properties relevant to pesticide performance, including those governing environmental behavior and efficacy.
• Molecular classification: Identifies candidate pesticides among structurally related molecules not previously classified as such.
• AI-guided optimization: Supports structural modification of identified candidates to eliminate undesirable properties such as toxicity indicators.
• Validated scope: The underlying DMS-ML and ML-only workflows have been applied to multiple molecular classes, including pesticides, in prior published work.

Technology readiness level

The core DMS-ML and ensemble stacking workflows have been demonstrated across multiple molecular classes, including pesticides, in peer-reviewed studies. WaterMine Innovation has continued to evolve these technologies since its founding in 2018 and is now seeking partners to advance the next validation phase. This phase will use the platform to screen for novel molecules exhibiting desirable pesticide characteristics and to apply AI-guided optimization to remove undesirable traits. The work is expected to take several months and requires partner support in ML model development, pesticide characteristic data, computational resources, and funding for key personnel.