Moving beyond MARCO

Abstract

The use of imaging systems in protein crystallisation means that the experimental setups no longer require manual inspection to determine the outcome of the trials. However, it leads to the problem of how best to find images which contain useful information about the crystallisation experiments. The adoption of a deeplearning approach in 2018 enabled a four-class machine classification system of the images to exceed human accuracy for the first time. Underpinning this was the creation of a labelled training set which came from a consortium of several different laboratories. The MARCO classification model does not have the same accuracy on local data as it does on images from the original test set; this can be somewhat mitigated by retraining the ML model and including local images. We have characterized the image data used in the original MARCO model, and performed extensive experiments to identify training settings most likely to enhance the local performance of a MARCO-dataset based ML classification model.

CHiMP: Deep Learning Tools Trained on Protein Crystallisation Micrographs to Enable Automation of Experiments

“…As part of a study which carried out a further investigation into the structure of the MARCO dataset, Rosa et al (47) trained a ResNet50 network on the data. They achieved a slight improvement in overall accuracy (94.63% vs the original MARCO score of 94.5%).…”

“…3 since this is the one used in our investigations. A detailed exploration of the MARCO dataset can be found in (47). Example images from each class can be seen in Figure 2.…”