DeepFold: enhancing protein structure prediction through optimized loss functions, improved template features, and re-optimized energy function.

MOTIVATION: Predicting protein structures with high accuracy is a critical challenge for the broad community of life sciences and industry. Despite progress made by deep neural networks like AlphaFold2, there is a need for further improvements in the quality of detailed structures, such as side-chains, along with protein backbone structures. RESULTS: Building upon the successes of AlphaFold2, the modifications we made include changing the losses of side-chain torsion angles and frame aligned point error, adding loss functions for side chain confidence and secondary structure prediction, and replacing template feature generation with a new alignment method based on conditional random fields. We also performed re-optimization by conformational space annealing using a molecular mechanics energy function which integrates the potential energies obtained from distogram and side-chain prediction. In the CASP15 blind test for single protein and domain modeling (109 domains), DeepFold ranked fourth among 132 groups with improvements in the details of the structure in terms of backbone, side-chain, and Molprobity. In terms of protein backbone accuracy, DeepFold achieved a median GDT-TS score of 88.64 compared with 85.88 of AlphaFold2. For TBM-easy/hard targets, DeepFold ranked at the top based on Z-scores for GDT-TS. This shows its practical value to the structural biology community, which demands highly accurate structures. In addition, a thorough analysis of 55 domains from 39 targets with publicly available structures indicates that DeepFold shows superior side-chain accuracy and Molprobity scores among the top-performing groups. AVAILABILITY AND IMPLEMENTATION: DeepFold tools are open-source software available at https://github.com/newtonjoo/deepfold.

Toward On-line Parameter Estimation of Concentric Tube Robots Using a Mechanics-based Kinematic Model.

Although existing mechanics-based models of concentric tube robots have been experimentally demonstrated to approximate the actual kinematics, determining accurate estimates of model parameters remains difficult due to the complex relationship between the parameters and available measurements. Further, because the mechanics-based models neglect some phenomena like friction, nonlinear elasticity, and cross section deformation, it is also not clear if model error is due to model simplification or to parameter estimation errors. The parameters of the superelastic materials used in these robots can be slowly time-varying, necessitating periodic re-estimation. This paper proposes a method for estimating the mechanics-based model parameters using an extended Kalman filter as a step toward on-line parameter estimation. Our methodology is validated through both simulation and experiments.

A minimum attention control law for ball catching.

Digital implementations of control laws typically involve discretization with respect to both time and space, and a control law that can achieve a task at coarser levels of discretization can be said to require less control attention, and also reduced implementation costs. One means of quantitatively capturing the attention of a control law is to measure the rate of change of the control with respect to changes in state and time. In this paper we present an attention-minimizing control law for ball catching and other target tracking tasks based on Brockett's attention criterion. We first highlight the connections between this attention criterion and some well-known principles from human motor control. Under the assumption that the optimal control law is the sum of a linear time-varying feedback term and a time-varying feedforward term, we derive an LQR-based minimum attention tracking control law that is stable, and obtained efficiently via a finite-dimensional optimization over the symmetric positive-definite matrices. Taking ball catching as our primary task, we perform numerical experiments comparing the performance of the various control strategies examined in the paper. Consistent with prevailing theories about human ball catching, our results exhibit several familiar features, e.g., the transition from open-loop to closed-loop control during the catching movement, and improved robustness to spatiotemporal discretization. The presented control laws are applicable to more general tracking problems that are subject to limited communication resources.

Kinematic feedback control laws for generating natural arm movements.

We propose a stochastic optimal feedback control law for generating natural robot arm motions. Our approach, inspired by the minimum variance principle of Harris and Wolpert (1998 Nature 394 780-4) and the optimal feedback control principles put forth by Todorov and Jordan (2002 Nature Neurosci. 5 1226-35) for explaining human movements, differs in two crucial respects: (i) the endpoint variance is minimized in joint space rather than Cartesian hand space, and (ii) we ignore the dynamics and instead consider only the second-order differential kinematics. The feedback control law generating the motions can be straightforwardly obtained by backward integration of a set of ordinary differential equations; these equations are obtained exactly, without any linear-quadratic approximations. The only parameters to be determined a priori are the variance scale factors, and for both the two-DOF planar arm and the seven-DOF spatial arm, a table of values is constructed based on the given initial and final arm configurations; these values are determined via an optimal fitting procedure, and consistent with existing findings about neuromuscular motor noise levels of human arm muscles. Experiments conducted with a two-link planar arm and a seven-DOF spatial arm verify that the trajectories generated by our feedback control law closely resemble human arm motions, in the sense of producing nearly straight-line hand trajectories, having bell-shaped velocity profiles, and satisfying Fitts Law.