Projects

Bacterial communities play key roles and plant health. Health and yield of crops, key for food security in light of climatic changes, is affected bybacterial small molecule metabolites, which signal to growth and defense pathways in the plant host. Their study through liquid chromatography -mass spectrometry (LC-MS) is seeing increased adoption, but remains underutilized due to challenges in metabolite identification, resultinterpretation and integration of data from diverse biological samples. Here, we propose the development of MS1-Tools, a machine learning modelto identify metabolites in general LC-MS datasets. By generating LC-MS datasets from bacterial species isolated from potato (Solanum tuberosum),we aim to optimize MS1-Tools for plant-bacterial compounds and discover novel growth-promoting molecules for this understudied, butindispensable food crop.

Identifying molecules from features in LC-MS has been limited by lack of reference measurements and difficulties in high throughput collection ofquality mass spectra. These challenges are exacerbated in plants due to their rich chemical repertoires. MS1-Tools uses probabilistic machinelearning to predict molecular properties (fingerprints) related to metabolites present in a sample. Here, we propose modern deep learning modulesas key innovations, including learning with sets and attention, which enable learning with incomplete data typical of LC-MS features. Together,molecular fingerprint prediction and deep learning promise comprehensive identification of molecules or molecular classes present in plants andplant-associated bacterial samples. Twenty-nine bacterial species isolated from potato fields and eight known plant-growth promoting bacteria willbe cultured and subjected to untargeted metabolomics.

These data will serve to train MS1-Tools and in turn chart bacterial metabolic repertoires, with focus on growth-promoting metabolite classes.Inoculation of promising monocultures into host plants will enable decoupling of the metabolic contribution from the plant host and bacterialsymbionts. This will provide an important resource to discover growth-promoting metabolites and their analogs, and unveil mechanisms for novel,marketable biostimulants and their effective producers.