Predicting microbial traits at scale through metagenomics and remote sensing

Project during my Science Undergraduate Laboratory Internship (SULI) at Lawrence Berkeley National Laboratory (LBNL)

My project focused on predicting microbial traits at scale using metagenomics and remote sensing, under the guidance of Dr. Nicola Falco and Dr. Ulas Karaoz. I  trained machine learning models and developed a pipeline that combines an XGBoost model with SHAP for interpretability, to model relationships between microbial functions and environmental predictors. The key aspect of this work was using remote sensing derived variables as proxies for microbial traits, enabling predictions beyond plot-level measurements. These inputs included hyperspectral-derived vegetation traits (such as leaf water content and leaf mass per area) and LiDAR-based landscape features like elevation, slope, and curvature. Through iterative training and feature selection, I found that remote sensing variables often outperformed traditional biogeochemical predictors. Combining both types of predictors improved performance for specific traits such as nitrogen transport and denitrification. This work demonstrated the potential of remote sensing to scale microbial trait prediction from localized samples to broader ecosystem-level analyses.

Read more about my internship experience here

Other outcomes from the internship:

• Poster presentation at LBNL towards the end of the internship (attached below)

• Co-authored an AGU Abstract with the research group