Machine learning is increasingly recognized as a promising technology in the biological, biomedical, and behavioral sciences. There can be no argument that this technique is incredibly successful in image recognition with immediate applications in diagnostics including electrophysiology, radiology, or pathology, where we have access to massive amounts of annotated data. However, machine learning often performs poorly in prognosis, especially when dealing with sparse data. This is a field where classical physics-based simulation seems to remain irreplaceable. In this review, we identify areas in the biomedical sciences where machine learning and multiscale modeling can mutually benefit from one another: Machine learning can integrate physics-based knowledge in the form of governing equations, boundary conditions, or constraints to manage ill-posted problems and robustly handle sparse and noisy data; multiscale modeling can integrate machine learning to create surrogate models, identify system dynamics and parameters, analyze sensitivities, and quantify uncertainty to bridge the scales and understand the emergence of function. With a view towards applications in the life sciences, we discuss the state of the art of combining machine learning and multiscale modeling, identify applications and opportunities, raise open questions, and address potential challenges and limitations. This review serves as introduction to a special issue on Uncertainty Quantification, Machine Learning, and Data-Driven Modeling of Biological Systems that will help identify current roadblocks and areas where computational mechanics, as a discipline, can play a significant role. We anticipate that it will stimulate discussion within the community of computational mechanics and reach out to other disciplines including mathematics, statistics, computer science, artificial intelligence, biomedicine, systems biology, and precision medicine to join forces towards creating robust and efficient models for biological systems.

We present a controls systems model of horizontal-plane head movements during perturbations of the trunk, which for the first time interfaces a model of the human head with neural feedback controllers representing the vestibulocollic (VCR) and the cervicocollic (CCR) reflexes. This model is homeomorphic such that model structure and parameters are drawn directly from anthropomorphic, biomechanical and physiological studies. Using control theory we analyzed the system model in the time and frequency domains, simulating neck movement responses to input perturbations of the trunk. Without reflex control, the head and neck system produced a second-order underdamped response with a 5.2 dB resonant peak at 2.1 Hz. Adding the CCR component to the system dampened the response by approximately 7%. Adding the VCR component dampened head oscillations by 75%. The VCR also improved low-frequency compensation by increasing the gain and phase lag, creating a phase minimum at 0.1 Hz and a phase peak at 1.1 Hz. Combining all three components (mechanics, VCR and CCR) linearly in the head and neck system reduced the amplitude of the resonant peak to 1.1 dB and increased the resonant frequency to 2.9 Hz. The closed loop results closely fit human data, and explain quantitatively the characteristic phase peak often observed.

The 3-dimensional angular vestibulo-ocular re¯exes (AVOR) elicited by rapid rotary head thrusts were studied in 17 subjects with unilateral Me Ânie Áre's disease before and 2±10 weeks after treatment with intratympanic gentamicin and in 13 subjects after surgical unilateral vestibular destruction (SUVD). Each head thrust was in the horizontal plane or in either diagonal plane of the vertical semicircular canals, so that each head thrust effectively stimulated only one pair of canals. The AVOR gains (eye velocity/head velocity during the 30 ms before peak head velocity) for the head thrusts exciting each individual canal were averaged and taken as a measure of the function of that canal. Prior to intratympanic gentamicin, gains for head thrusts that excited canals on the affected side were 0.91 0.20 (horizontal canal, HC), 0.78 0.20 (anterior canal, AC), and 0.83 0.10 (posterior canal, PC). The asymmetries between these gain values and those for head thrusts that excited the contralateral canals were <2%. In contrast, caloric asymmetries averaged 40% 32%. Intratympanic gentamicin resulted in decreased gains attributable to each canal on the treated side: 0.40 0.12 (HC), 0.35 0.14 (AC), 0.31 0.14 (PC) (p < 0.01). However, the gains attributable to contralateral canals dropped only slightly, resulting in marked asymmetries between gains for excitation of ipsilateral canals versus their contralateral mates: HC: 34% 12%, AC: 24% 25%, and PC: 42% 13%. There was no difference in the AVOR gain for excitation of the ipsilateral HC after gentamicin in patients who received a single intratympanic injection (0.39 0.11, n = 12) in comparison to those who received 2±3 injections (0.42 0.15, n = 5, p = 0.7). Gain decreases attributed to the gentamicintreated HC and AC were not as severe as those observed after SUVD. This ®nding suggests that intratympanic gentamicin causes a partial vestibular lesion that may involve preservation of spontaneous discharge and/or rotational sensitivity of afferents.

The complexities of modern biomedicine are rapidly increasing. Thus, modeling and simulation have become increasingly important as a strategy to understand and predict the trajectory of pathophysiology, disease genesis, and disease spread in support of clinical and policy decisions. In such cases, inappropriate or ill-placed trust in the model and simulation outcomes may result in negative outcomes, and hence illustrate the need to formalize the execution and communication of modeling and simulation practices. Although verification and validation have been generally accepted as significant components of a model’s credibility, they cannot be assumed to equate to a holistic credible practice, which includes activities that can impact comprehension and in-depth examination inherent in the development and reuse of the models. For the past several years, the Committee on Credible Practice of Modeling and Simulation in Healthcare, an interdisciplinary group seeded from a U.S. interagency initiative, has worked to codify best practices. Here, we provide Ten Rules for credible practice of modeling and simulation in healthcare developed from a comparative analysis by the Committee’s multidisciplinary membership, followed by a large stakeholder community survey. These rules establish a unified conceptual framework for modeling and simulation design, implementation, evaluation, dissemination and usage across the modeling and simulation life-cycle. While biomedical science and clinical care domains have somewhat different requirements and expectations for credible practice, our study converged on rules that would be useful across a broad swath of model types. In brief, the rules are: (1) Define context clearly. (2) Use contextually appropriate data. (3) Evaluate within context. (4) List limitations explicitly. (5) Use version control. (6) Document appropriately. (7) Disseminate broadly. (8) Get independent reviews. (9) Test competing implementations. (10) Conform to standards. Although some of these are common sense guidelines, we have found that many are often missed or misconstrued, even by seasoned practitioners. Computational models are already widely used in basic science to generate new biomedical knowledge. As they penetrate clinical care and healthcare policy, contributing to personalized and precision medicine, clinical safety will require established guidelines for the credible practice of modeling and simulation in healthcare.