A new paper that we worked on has been published at JASMS, titled: Mass Spectrometry Imaging for Spatial Ingredient Classification in Plant-Based Food. Ever wondered how we can apply ML on MSI data for classification of constituents? In this paper we show three approaches to do this. Firstly an unsupervised approach to figure out the hetereogeneity or adulteration of your samples. Secondly a semi-supervised approach to train and identify ingredients in plant-based burger. Finally, a supervised approach to figure out important peaks that can separate between different samples or, in our cases, different burgers. Find our paper by clicking here
I am excited to announce to that our tutorial paper on how to apply Random Forest classifiers to ToF-SIMS imaging data has been published at JASMS and can be found here . Analysis of ToF-SIMS datasets can be quite challenging due to the lack or low intensity of molecular ions and the abundance of fragments leading to highly dimensional and highly collinear data. While Random Forest models have some robustness against these problems, we show that we can improve the performance of our classifier model by setting collinearity thresholds and removing redundant features. At the same time this leads to a cheaper & simpler model to train with less overfitting. What’s there not to like?