Data reduction for X-ray serial crystallography using machine learning

Abstract

Serial crystallography experiments produce massive amounts of experimental data. Yet in spite of these large-scale data sets, only a small percentage of the data are useful for downstream analysis. Thus, it is essential to differentiate reliably between acceptable data (hits) and unacceptable data (misses). To this end, a novel pipeline is proposed to categorize the data, which extracts features from the images, summarizes these features with the `bag of visual words' method and then classifies the images using machine learning. In addition, a novel study of various feature extractors and machine learning classifiers is presented, with the aim of finding the best feature extractor and machine learning classifier for serial crystallography data. The study reveals that the oriented FAST and rotated BRIEF (ORB) feature extractor with a multilayer perceptron classifier gives the best results. Finally, the ORB feature extractor with multilayer perceptron is evaluated on various data sets including both synthetic and experimental data, demonstrating superior performance compared with other feature extractors and classifiers.

Explainable machine learning for diffraction patterns

“…DeepFreak (GLCM + random forest) (Souza et al, 2019) 98.4 DeepFreak (GLCM + support vector machine) (Souza et al, 2019) 97.6 ORB + MLP (Rahmani et al, 2023) 97.5 DeepFreak (Souza et al, 2019) 98.8 AlexNet (our implementation) 98.1 ResNet-101 (our implementation) 98.3 which are useful for downstream tasks such as indexing. On the other hand, deep learning methods reliably differentiate between hit and miss classes.…”

“…Previous work involving machine learning has used both synthetic and experimental data sets (Ke et al, 2018;Souza et al, 2019;Rahmani et al, 2023). Thus, we selected data sets to visualize CNN representations along with the parts respon-sible for a certain prediction.…”

“…With these advancements, the crystallography community has also made use of machine learning for various applications (Sullivan et al, 2019;Park et al, 2017;Ryan et al, 2018;Wang et al, 2020). Specifically, the serial crystallography community has experimented with these methods to achieve data reduction (Becker & Streit, 2014;Ke et al, 2018;Souza et al, 2019;Rahmani et al, 2023;Chen et al, 2021). Machine learning, or more specifically deep learning methods including convolutional neural networks (CNNs), encode experimental data to classify it into hit or miss categories.…”

Deep residual networks for crystallography trained on synthetic data

“…refinement, but with the addition of new AI programs (Ke et al, 2018;Rahmani et al, 2023) to screen for diffraction, and our present body of work that uses AI to characterize diffraction, we are edging away from requiring user interactions for serial data processing.…”

“…These 'misses' (which comprise significant percentages of data collected using high-flow-rate injector methods) could then be excluded from processing and/or recording to disk to free up computing resources. More recently, in Rahmani et al (2023) various dimensionalityreduction algorithms have been used to convert diffraction data into a set of features suitable for training a machinelearning classifier to automatically detect whether experimental images contained diffraction.…”