Gaze entropy metrics for mental workload estimation are heterogenous during hands-off level 2 automation.

As the level of vehicle automation increases, drivers are more likely to engage in non-driving related tasks which take their hands, eyes, and/or mind away from the driving task. Consequently, there has been increased interest in creating Driver Monitoring Systems (DMS) that are valid and reliable for detecting elements of driver state. Workload is one element of driver state that has remained elusive within the literature. Whilst there has been promising work in estimating mental workload using gaze-based metrics, the literature has placed too much emphasis on point estimate differences. Whilst these are useful for establishing whether effects exist, they ignore the inherent variability within individuals and between different drivers. The current work builds on this by using a Bayesian distributional modelling approach to quantify the within and between participants variability in Information Theoretical gaze metrics. Drivers (N = 38) undertook two experimental drives in hands-off Level 2 automation with their hands and feet away from operational controls. During both drives, their priority was to monitor the road before a critical takeover. During one drive participants had to complete a secondary cognitive task (2-back) during the hands-off Level 2 automation. Changes in Stationary Gaze Entropy and Gaze Transition Entropy were assessed for conditions with and without the 2-back to investigate whether consistent differences between workload conditions could be found across the sample. Stationary Gaze Entropy proved a reliable indicator of mental workload; 92 % of the population were predicted to show a decrease when completing 2-back during hands-off Level 2 automated driving. Conversely, Gaze Transition Entropy showed substantial heterogeneity; only 66 % of the population were predicted to have similar decreases. Furthermore, age was a strong predictor of the heterogeneity of the average causal effect that high mental workload had on eye movements. These results indicate that, whilst certain elements of Information Theoretic metrics can be used to estimate mental workload by DMS, future research needs to focus on the heterogeneity of these processes. Understanding this heterogeneity has important implications toward the design of future DMS and thus the safety of drivers using automated vehicle functions. It must be ensured that metrics used to detect mental workload are valid (accurately detecting a particular driver state) as well as reliable (consistently detecting this driver state across a population).

Who goes first? A distributed simulator study of vehicle-pedestrian interaction.

One of the current challenges of automation is to have highly automated vehicles (HAVs) that communicate effectively with pedestrians and react to changes in pedestrian behaviour, to promote more trustable HAVs. However, the details of how human drivers and pedestrians interact at unsignalised crossings remain poorly understood. We addressed some aspects of this challenge by replicating vehicle-pedestrian interactions in a safe and controlled virtual environment by connecting a high fidelity motion-based driving simulator to a CAVE-based pedestrian lab in which 64 participants (32 pairs of one driver and one pedestrian) interacted with each other under different scenarios. The controlled setting helped us study the causal role of kinematics and priority rules on interaction outcome and behaviour, something that is not possible in naturalistic studies. We also found that kinematic cues played a stronger role than psychological traits like sensation seeking and social value orientation in determining whether the pedestrian or driver passed first at unmarked crossings. One main contribution of this study is our experimental paradigm, which permitted repeated observation of crossing interactions by each driver-pedestrian participant pair, yielding behaviours which were qualitatively in line with observations from naturalistic studies.

Drivers' Evaluation of Different Automated Driving Styles: Is It Both Comfortable and Natural?

OBJECTIVE: This study investigated users' subjective evaluation of three highly automated driving styles, in terms of comfort and naturalness, when negotiating a UK road in a high-fidelity, motion-based, driving simulator. BACKGROUND: Comfort and naturalness play an important role in contributing to users' acceptance and trust of automated vehicles (AVs), although not much is understood about the types of driving style which are considered comfortable or natural. METHOD: A driving simulator study, simulating roads with different road geometries and speed limits, was conducted. Twenty-four participants experienced three highly automated driving styles, two of which were recordings from human drivers, and the other was based on a machine learning (ML) algorithm, termed Defensive, Aggressive, and Turner, respectively. Participants evaluated comfort or naturalness of each driving style, for each road segment, and completed a Sensation Seeking questionnaire, which assessed their risk-taking propensity. RESULTS: Participants regarded both human-like driving styles as more comfortable and natural, compared with the less human-like, ML-based, driving controller. Particularly, between the two human-like controllers, the Defensive style was considered more comfortable, especially for the more challenging road environments. Differences in preference for controller by driver trait were also observed, with the Aggressive driving style evaluated as more natural by the high sensation seekers. CONCLUSION: Participants were able to distinguish between human- and machine-like AV controllers. A range of psychological concepts must be considered for the subjective evaluation of controllers. APPLICATION: Insights into how different driver groups evaluate automated vehicle controllers are important in designing more acceptable systems.

Fluctuating Finite Element Analysis (FFEA): A continuum mechanics software tool for mesoscale simulation of biomolecules.

Fluctuating Finite Element Analysis (FFEA) is a software package designed to perform continuum mechanics simulations of proteins and other globular macromolecules. It combines conventional finite element methods with stochastic thermal noise, and is appropriate for simulations of large proteins and protein complexes at the mesoscale (length-scales in the range of 5 nm to 1 µm), where there is currently a paucity of modelling tools. It requires 3D volumetric information as input, which can be low resolution structural information such as cryo-electron tomography (cryo-ET) maps or much higher resolution atomistic co-ordinates from which volumetric information can be extracted. In this article we introduce our open source software package for performing FFEA simulations which we have released under a GPLv3 license. The software package includes a C ++ implementation of FFEA, together with tools to assist the user to set up the system from Electron Microscopy Data Bank (EMDB) or Protein Data Bank (PDB) data files. We also provide a PyMOL plugin to perform basic visualisation and additional Python tools for the analysis of FFEA simulation trajectories. This manuscript provides a basic background to the FFEA method, describing the implementation of the core mechanical model and how intermolecular interactions and the solvent environment are included within this framework. We provide prospective FFEA users with a practical overview of how to set up an FFEA simulation with reference to our publicly available online tutorials and manuals that accompany this first release of the package.

Blind prediction of interfacial water positions in CAPRI.

We report the first assessment of blind predictions of water positions at protein-protein interfaces, performed as part of the critical assessment of predicted interactions (CAPRI) community-wide experiment. Groups submitting docking predictions for the complex of the DNase domain of colicin E2 and Im2 immunity protein (CAPRI Target 47), were invited to predict the positions of interfacial water molecules using the method of their choice. The predictions-20 groups submitted a total of 195 models-were assessed by measuring the recall fraction of water-mediated protein contacts. Of the 176 high- or medium-quality docking models-a very good docking performance per se-only 44% had a recall fraction above 0.3, and a mere 6% above 0.5. The actual water positions were in general predicted to an accuracy level no better than 1.5 Å, and even in good models about half of the contacts represented false positives. This notwithstanding, three hotspot interface water positions were quite well predicted, and so was one of the water positions that is believed to stabilize the loop that confers specificity in these complexes. Overall the best interface water predictions was achieved by groups that also produced high-quality docking models, indicating that accurate modelling of the protein portion is a determinant factor. The use of established molecular mechanics force fields, coupled to sampling and optimization procedures also seemed to confer an advantage. Insights gained from this analysis should help improve the prediction of protein-water interactions and their role in stabilizing protein complexes.

Expanding the frontiers of protein-protein modeling: from docking and scoring to binding affinity predictions and other challenges.

In addition to protein-protein docking, this CAPRI edition included new challenges, like protein-water and protein-sugar interactions, or the prediction of binding affinities and ΔΔG changes upon mutation. Regarding the standard protein-protein docking cases, our approach, mostly based on the pyDock scheme, submitted correct models as predictors and as scorers for 67% and 57% of the evaluated targets, respectively. In this edition, available information on known interface residues hardly made any difference for our predictions. In one of the targets, the inclusion of available experimental small-angle X-ray scattering (SAXS) data using our pyDockSAXS approach slightly improved the predictions. In addition to the standard protein-protein docking assessment, new challenges were proposed. One of the new problems was predicting the position of the interface water molecules, for which we submitted models with 20% and 43% of the water-mediated native contacts predicted as predictors and scorers, respectively. Another new problem was the prediction of protein-carbohydrate binding, where our submitted model was very close to being acceptable. A set of targets were related to the prediction of binding affinities, in which our pyDock scheme was able to discriminate between natural and designed complexes with area under the curve = 83%. It was also proposed to estimate the effect of point mutations on binding affinity. Our approach, based on machine learning methods, showed high rates of correctly classified mutations for all cases. The overall results were highly rewarding, and show that the field is ready to move forward and face new interesting challenges in interactomics.

Efficient Relaxation of Protein-Protein Interfaces by Discrete Molecular Dynamics Simulations.

Protein-protein interactions are responsible for the transfer of information inside the cell and represent one of the most interesting research fields in structural biology. Unfortunately, after decades of intense research, experimental approaches still have difficulties in providing 3D structures for the hundreds of thousands of interactions formed between the different proteins in a living organism. The use of theoretical approaches like docking aims to complement experimental efforts to represent the structure of the protein interactome. However, we cannot ignore that current methods have limitations due to problems of sampling of the protein-protein conformational space and the lack of accuracy of available force fields. Cases that are especially difficult for prediction are those in which complex formation implies a non-negligible change in the conformation of the interacting proteins, i.e., those cases where protein flexibility plays a key role in protein-protein docking. In this work, we present a new approach to treat flexibility in docking by global structural relaxation based on ultrafast discrete molecular dynamics. On a standard benchmark of protein complexes, the method provides a general improvement over the results obtained by rigid docking. The method is especially efficient in cases with large conformational changes upon binding, in which structure relaxation with discrete molecular dynamics leads to a predictive success rate double that obtained with state-of-the-art rigid-body docking.

pyDockCG: new coarse-grained potential for protein-protein docking.

Protein-protein interactions are fundamental for the majority of biological processes, so their structural, functional, and energetic characterization is of enormous biotechnological and therapeutic interest. In recent years, a variety of computational docking approaches to the structural prediction of protein-protein complexes have been reported, with encouraging results. However, a major bottleneck is found in cases with conformational movements upon binding, for which docking algorithms have to be extended beyond the rigid-body framework by introducing flexibility. Given the high computational cost of flexible docking, coarse-grained models offer an efficient alternative to full-atom descriptions. This work describes pyDockCG, a new coarse-grained potential for protein-protein docking scoring and refinement, based on the known UNRES model for polypeptide chains. The main novelty is the inclusion of two new terms accounting for the Coulomb electrostatics and the solvation energy. The latter has been devised by adapting the EEF1 model to the coarse-grained approach, with optimal parameters for protein-protein docking. The coarse-grained potential yielded highly similar values to the full-atom scoring function pyDock when applied to the rigid body docking sets, but at much lower computational cost. This efficiency makes it suitable for the treatment of flexibility during docking.

Optimization of pyDock for the new CAPRI challenges: Docking of homology-based models, domain-domain assembly and protein-RNA binding.

We describe here our results in the last CAPRI edition. We have participated in all targets, both as predictors and as scorers, using our pyDock docking methodology. The new challenges (homology-based modeling of the interacting subunits, domain-domain assembling, and protein-RNA interactions) have pushed our computer tools to the limits and have encouraged us to devise new docking approaches. Overall, the results have been quite successful, in line with previous editions, especially considering the high difficulty of some of the targets. Our docking approaches succeeded in five targets as predictors or as scorers (T29, T34, T35, T41, and T42). Moreover, with the inclusion of available information on the residues expected to be involved in the interaction, our protocol would have also succeeded in two additional cases (T32 and T40). In the remaining targets (except T37), results were equally poor for most of the groups. We submitted the best model (in ligand RMSD) among scorers for the unbound-bound target T29, the second best model among scorers for the protein-RNA target T34, and the only correct model among predictors for the domain assembly target T35. In summary, our excellent results for the new proposed challenges in this CAPRI edition showed the limitations and applicability of our approaches and encouraged us to continue developing methodologies for automated biomolecular docking.

Protein docking by Rotation-Based Uniform Sampling (RotBUS) with fast computing of intermolecular contact distance and residue desolvation.

BACKGROUND: Protein-protein interactions are fundamental for the majority of cellular processes and their study is of enormous biotechnological and therapeutic interest. In recent years, a variety of computational approaches to the protein-protein docking problem have been reported, with encouraging results. Most of the currently available protein-protein docking algorithms are composed of two clearly defined parts: the sampling of the rotational and translational space of the interacting molecules, and the scoring and clustering of the resulting orientations. Although this kind of strategy has shown some of the most successful results in the CAPRI blind test http://www.ebi.ac.uk/msd-srv/capri, more efforts need to be applied. Thus, the sampling protocol should generate a pool of conformations that include a sufficient number of near-native ones, while the scoring function should discriminate between near-native and non-near-native proposed conformations. On the other hand, protocols to efficiently include full flexibility on the protein structures are increasingly needed. RESULTS: In these work we present new computational tools for protein-protein docking. We describe here the RotBUS (Rotation-Based Uniform Sampling) method to generate uniformly distributed sets of rigid-body docking poses, with a new fast calculation of the optimal contacting distance between molecules. We have tested the method on a standard benchmark of unbound structures and we can find near-native solutions in 100% of the cases. After applying a new fast filtering scheme based on residue-based desolvation, in combination with FTDock plus pyDock scoring, near-native solutions are found with rank

Structural prediction of protein-RNA interaction by computational docking with propensity-based statistical potentials.

Despite the importance of protein-RNA interactions in the cellular context, the number of available protein-RNA complex structures is still much lower than those of other biomolecules. As a consequence, few computational studies have been addressed towards protein-RNA complexes, and to our knowledge, no systematic benchmarking of protein-RNA docking has been reported. In this study we have extracted new pairwise residue-ribonucleotide interface propensities for protein-RNA, which can be used as statistical potentials for scoring of protein-RNA docking poses. We show here a new protein-RNA docking approach based on FTDock generation of rigid-body docking poses, which are later scored by these statistical residue-ribonucleotide potentials. The method has been successfully benchmarked in a set of 12 protein-RNA cases. The results show that FTDock is able to generate near-native solutions in more than half of the cases, and that it can rank near-native solutions significantly above random. In practically all these cases, our propensity-based scoring helps to improve the docking results, finding a near-native solution within rank 100 in 43% of them. In a remarkable case, the near-native solution was ranked 1 after the propensity-based scoring. Other previously described propensity potentials can also be used for scoring, with slightly worse performance. This new protein-RNA docking protocol permits a fast scoring of rigid-body docking poses in order to select a small number of docking orientations, which can be later evaluated with more sophisticated energy-based scoring functions.

Present and future challenges and limitations in protein-protein docking.

The study of protein-protein interactions that are involved in essential life processes can largely benefit from the recent upraising of computational docking approaches. Predicting the structure of a protein-protein complex from their separate components is still a highly challenging task, but the field is rapidly improving. Recent advances in sampling algorithms and rigid-body scoring functions allow to produce, at least for some cases, high quality docking models that are perfectly suitable for biological and functional annotations, as it has been shown in the CAPRI blind tests. However, important challenges still remain in docking prediction. For example, in cases with significant mobility, such as multidomain proteins, fully unrestricted rigid-body docking approaches are clearly insufficient so they need to be combined with restraints derived from domain-domain linker residues, evolutionary information, or binding site predictions. Other challenging cases are weak or transient interactions, such as those between proteins involved in electron transfer, where the existence of alternative bound orientations and encounter complexes complicates the binding energy landscape. Docking methods also struggle when using in silico structural models for the interacting subunits. Bringing these challenges to a practical point of view, we have studied here the limitations of our docking and energy-based scoring approach, and have analyzed different parameters to overcome the limitations and improve the docking performance. For that, we have used the standard benchmark and some practical cases from CAPRI. Based on these results, we have devised a protocol to estimate the success of a given docking run.

Prediction and scoring of docking poses with pyDock.

The two previous CAPRI experiments showed the success of our rigid-body and refinement approach. For this third edition of CAPRI, we have used a new faster protocol called pyDock, which uses electrostatics and desolvation energy to score docking poses generated with FFT-based algorithms. In target T24 (unbound/model), our best prediction had the highest value of fraction of native contacts (40%) among all participants, although it was not considered as acceptable by the CAPRI criteria. In target T25 (unbound/bound), we submitted a model with medium quality. In target T26 (unbound/unbound), we did not submit any acceptable model (but we would have submitted acceptable predictions if we had included available mutational information about the binding site). For targets T27 (unbound/unbound) and T28 (homo-dimer using model), nobody (including us) submitted any acceptable model. Intriguingly, the crystal structure of target T27 shows an alternative interface that correlates with available biological data (we would have submitted acceptable predictions if we had included this). We also participated in all targets of the SCORERS experiment, with at least acceptable accuracy in all valid cases. We submitted two medium and four acceptable scoring models of T25. Using additional distance restraints (from mutational data), we had two medium and two acceptable scoring models of T26. For target T27, we submitted two acceptable scoring models of the alternative interface in the crystal structure. In summary, CAPRI showed the excellent capabilities of pyDock in identifying near-native docking poses.