Development and internal validation of machine learning-based models and external validation of existing risk scores for outcome prediction in patients with ischaemic stroke

Aims

We developed new machine learning (ML) models and externally validated existing statistical models [ischaemic stroke predictive risk score (iScore) and totalled health risks in vascular events (THRIVE) scores] for predicting the composite of recurrent stroke or all-cause mortality at 90 days and at 3 years after hospitalization for first acute ischaemic stroke (AIS).

Methods and results

In adults hospitalized with AIS from January 2005 to November 2016, with follow-up until November 2019, we developed three ML models [random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBOOST)] and externally validated the iScore and THRIVE scores for predicting the composite outcomes after AIS hospitalization, using data from 721 patients and 90 potential predictor variables. At 90 days and 3 years, 11 and 34% of patients, respectively, reached the composite outcome. For the 90-day prediction, the area under the receiver operating characteristic curve (AUC) was 0.779 for RF, 0.771 for SVM, 0.772 for XGBOOST, 0.720 for iScore, and 0.664 for THRIVE. For 3-year prediction, the AUC was 0.743 for RF, 0.777 for SVM, 0.773 for XGBOOST, 0.710 for iScore, and 0.675 for THRIVE.

Conclusion

The study provided three ML-based predictive models that achieved good discrimination and clinical usefulness in outcome prediction after AIS and broadened the application of the iScore and THRIVE scoring system for long-term outcome prediction. Our findings warrant comparative analyses of ML and existing statistical method-based risk prediction tools for outcome prediction after AIS in new data sets.

Serial femtosecond and serial synchrotron crystallography can yield data of equivalent quality: A systematic comparison

For the two proteins myoglobin and fluoroacetate dehalogenase, we present a systematic comparison of crystallographic diffraction data collected by serial femtosecond (SFX) and serial synchrotron crystallography (SSX). To maximize comparability, we used the same batch of micron-sized crystals, the same sample delivery device, and the same data analysis software. Overall figures of merit indicate that the data of both radiation sources are of equivalent quality. For both proteins, reasonable data statistics can be obtained with approximately 5000 room-temperature diffraction images irrespective of the radiation source. The direct comparability of SSX and SFX data indicates that the quality of diffraction data obtained from these samples is linked to the properties of the crystals rather than to the radiation source. Therefore, for other systems with similar properties, time-resolved experiments can be conducted at the radiation source that best matches the desired time resolution.

Non-Contact Universal Sample Presentation for Room Temperature Macromolecular Crystallography Using Acoustic Levitation

Macromolecular Crystallography is a powerful and valuable technique to assess protein structures. Samples are commonly cryogenically cooled to minimise radiation damage effects from the X-ray beam, but low temperatures hinder normal protein functions and this procedure can introduce structural artefacts. Previous experiments utilising acoustic levitation for beamline science have focused on Langevin horns which deliver significant power to the confined droplet and are complex to set up accurately. In this work, the low power, portable TinyLev acoustic levitation system is used in combination with an approach to dispense and contain droplets, free of physical sample support to aid protein crystallography experiments. This method facilitates efficient X-ray data acquisition in ambient conditions compatible with dynamic studies. Levitated samples remain free of interference from fixed sample mounts, receive negligible heating, do not suffer significant evaporation and since the system occupies a small volume, can be readily installed at other light sources.

Dynamic adaptive X-ray optics. Part II. High-speed piezoelectric bimorph deformable Kirkpatrick-Baez mirrors for rapid variation of the 2D size and shape of X-ray beams

The tangential curvature of actively bent X-ray mirrors at synchrotron radiation and X-ray free-electron laser (XFEL) facilities is typically only changed every few hours or even days. This operation can take tens of minutes for active optics with multiple bending actuators and often requires expert guidance using in situ monitoring devices. Hence, the dynamic performance of active X-ray optics for synchrotron beamlines has historically not been exploited. This is in stark contrast to many other scientific fields. However, many areas of synchrotron radiation and XFEL science, including macromolecular crystallography, could greatly benefit from the ability to change the size and shape of the X-ray beam rapidly and continuously. The advantages of this innovative approach are twofold: a large reduction in the dead time required to change the size of the X-ray beam for different-sized samples and the possibility of making multiple changes to the beam during the measurement of a single sample. In the preceding paper [Part I; Alcock, Nistea, Signorato & Sawhney (2019), J. Synchrotron Rad. 26, 36-44], which accompanies this article, high-speed visible-light Fizeau interferometry was used to identify the factors which influence the dynamic bending behaviour of piezoelectric bimorph deformable X-ray mirrors. Building upon this ex situ metrology study, provided here is the first synchrotron radiation beamline implementation of high-speed adaptive X-ray optics using two bimorphs operating as a Kirkpatrick-Baez pair. With optimized substrates, novel opto-mechanical holders and a next-generation high-voltage power supply, the size of an X-ray beam was rapidly and repeatedly switched in <10 s. Of equal importance, it is also shown that compensation of piezoelectric creep ensures that the X-ray beam size remains stable for more than 1 h after making a major change. The era of high-speed adaptive X-ray optics for synchrotron radiation and XFEL beamlines has begun.

An acoustic on-chip goniometer for room temperature macromolecular crystallography

This paper describes the design, development and successful use of an on-chip goniometer for room-temperature macromolecular crystallography via acoustically induced rotations. We present for the first time a low cost, rate-tunable, acoustic actuator for gradual in-fluid sample reorientation about varying axes and its utilisation for protein structure determination on a synchrotron beamline. The device enables the efficient collection of diffraction data via a rotation method from a sample within a surface confined droplet. This method facilitates efficient macromolecular structural data acquisition in fluid environments for dynamical studies.

Time zero determination for FEL pump-probe studies based on ultrafast melting of bismuth

A common challenge for pump-probe studies of structural dynamics at X-ray free-electron lasers (XFELs) is the determination of time zero (T₀)-the time an optical pulse (e.g., an optical laser) arrives coincidently with the probe pulse (e.g., a XFEL pulse) at the sample position. In some cases, T₀ might be extracted from the structural dynamics of the sample's observed response itself, but generally, an independent robust method is required or would be superior to the inferred determination of T₀. In this paper, we present how the structural dynamics in ultrafast melting of bismuth can be exploited for a quickly performed, reliable and accurate determination of T₀ with a precision below 20 fs and an overall experimental accuracy of 50 fs to 150 fs (estimated). Our approach is potentially useful and applicable for fixed-target XFEL experiments, such as serial femtosecond crystallography, utilizing an optical pump pulse in the ultraviolet to near infrared spectral range and a pixelated 2D photon detector for recording crystallographic diffraction patterns in transmission geometry. In comparison to many other suitable approaches, our method is fairly independent of the pumping wavelength (UV-IR) as well as of the X-ray energy and offers a favorable signal contrast. The technique is exploitable not only for the determination of temporal characteristics of the experiment at the interaction point but also for investigating important conditions affecting experimental control such as spatial overlap and beam spot sizes.

Molecularly Resolved Protein Electromechanical Properties

Previous work has shown that protein molecules can be trapped between the conductive surfaces presented by a metal-coated AFM probe and an underlying planar substrate where their molecule-specific conductance characteristics can be assayed. Herein, we demonstrate that transport across such a derived metal-protein-electrode junction falls within three, pressure-dependent, regimes and, further, that pressure-dependent conductance can be utilized in analyzing temporal variations of protein fold. Specifically, the electronic and mechanical properties of the metalloprotein azurin have been characterized under conditions of anisotropic vertical compression through the use of a conducting atomic force microscope (CP-AFM). By utilizing the ability of azurin to chemically self-assemble on the gold surface presented either by the apex of a suitably coated AFM probe or a planar metallic surface, molecular-level transport characteristics are assayable. Under conditions of low force, typically less than 2 nN, the weak physical and electronic coupling between the protein and the conducting contacts impedes tunneling and leads to charge buildup followed by dielectric breakdown. At slightly increased force, 3-5 nN, the copper protein exhibits temporal electron occupation with observable negative differential resistance, while the redox-inactive zinc mutant does not. At imposed loads greater than 5 nN, appreciable electron tunneling can be detected even at low bias for both the redox-active and -inactive species. Dynamic current-voltage characteristics have been recorded and are well-described by a modified Simmons tunneling model. Subsequent analyses enable the electron tunneling barrier height and barrier length to be determined under conditions of quantified vertical stress. The variance observed describes, in essence, the protein's mechanical properties within the confines of the tunnel junction.