Are Metrics Enough? Guidelines for Communicating and Visualizing Predictive Models to Subject Matter Experts.

Presenting a predictive model's performance is a communication bottleneck that threatens collaborations between data scientists and subject matter experts. Accuracy and error metrics alone fail to tell the whole story of a model - its risks, strengths, and limitations - making it difficult for subject matter experts to feel confident in their decision to use a model. As a result, models may fail in unexpected ways or go entirely unused, as subject matter experts disregard poorly presented models in favor of familiar, yet arguably substandard methods. In this paper, we describe an iterative study conducted with both subject matter experts and data scientists to understand the gaps in communication between these two groups. We find that, while the two groups share common goals of understanding the data and predictions of the model, friction can stem from unfamiliar terms, metrics, and visualizations - limiting the transfer of knowledge to SMEs and discouraging clarifying questions being asked during presentations. Based on our findings, we derive a set of communication guidelines that use visualization as a common medium for communicating the strengths and weaknesses of a model. We provide a demonstration of our guidelines in a regression modeling scenario and elicit feedback on their use from subject matter experts. From our demonstration, subject matter experts were more comfortable discussing a model's performance, more aware of the trade-offs for the presented model, and better equipped to assess the model's risks - ultimately informing and contextualizing the model's use beyond text and numbers.

Leading a Digital Transformation in the Pharmaceutical Industry: Reimagining the Way We Work in Global Drug Development.

We are experiencing seminal times in computing that seem to define a fourth industrial revolution. This may fundamentally change the way we live, work, and relate to one another. Embracing data and digital information is a top priority for most industries these days, and Life Sciences is no exception. The pharmaceutical industry in particular is fundamentally a data-driven business. Inspired by a desire to "Go Big on Data," we developed a strategic roadmap defining a digital transformation to reimagine the way we work in Novartis Global Drug Development, leveraging data science to generate and inject actionable insights into our best practices. We launched a program called Nerve Live, and built a state-of-the-art data and analytics platform to harness past and present operational data, providing access to decades of drug development "experience" buried across multiple sources. The platform enabled the systematic application of machine learning and predictive analytics to generate "intelligence": new insights across multiple functional areas. To action the insights and create "value," we crafted skillfully designed end-user applications for domain experts to plan, track, predict, compare and monitor domain activities, optimize costs, and maximize quality. Today, the Nerve Live program enables insights-driven decision making at scale, unlocking productivity, and providing transparency across the Novartis Global Drug Development organization and beyond. We identified three main drivers making the Nerve Live program successful and enabling the associated digital transformation to flourish. We discuss the challenges, highlight the benefits, and see the importance of leading the way to become future proof.

Understanding pharmacokinetics using realistic computational models of fluid dynamics: biosimulation of drug distribution within the CSF space for intrathecal drugs.

We introduce how biophysical modeling in pharmaceutical research and development, combining physiological observations at the tissue, organ and system level with selected drug physiochemical properties, may contribute to a greater and non-intuitive understanding of drug pharmacokinetics and therapeutic design. Based on rich first-principle knowledge combined with experimental data at both conception and calibration stages, and leveraging our insights on disease processes and drug pharmacology, biophysical modeling may provide a novel and unique opportunity to interactively characterize detailed drug transport, distribution, and subsequent therapeutic effects. This innovative approach is exemplified through a three-dimensional (3D) computational fluid dynamics model of the spinal canal motivated by questions arising during pharmaceutical development of one molecular therapy for spinal cord injury. The model was based on actual geometry reconstructed from magnetic resonance imaging data subsequently transformed in a parametric 3D geometry and a corresponding finite-volume representation. With dynamics controlled by transient Navier-Stokes equations, the model was implemented in a commercial multi-physics software environment established in the automotive and aerospace industries. While predictions were performed in silico, the underlying biophysical models relied on multiple sources of experimental data and knowledge from scientific literature. The results have provided insights into the primary factors that can influence the intrathecal distribution of drug after lumbar administration. This example illustrates how the approach connects the causal chain underlying drug distribution, starting with the technical aspect of drug delivery systems, through physiology-driven drug transport, then eventually linking to tissue penetration, binding, residence, and ultimately clearance. Currently supporting our drug development projects with an improved understanding of systems physiology, biophysical models are being increasingly used to characterize drug transport and distribution in human tissues where pharmacokinetic measurements are difficult or impossible to perform. Importantly, biophysical models can describe emergent properties of a system, i.e. properties not identifiable through the study of the system's components taken in isolation.

The electroencephalographic fingerprint of sleep is genetically determined: a twin study.

Humans have an individual profile of the electroencephalographic power spectra at the 8 to 16 Hz frequency during non-rapid eye movement sleep that is stable over time and resistant to experimental perturbations. We tested the hypothesis that this electroencephalographic "fingerprint" is genetically determined, by recording 40 monozygotic and dizygotic twins during baseline and recovery sleep after prolonged wakefulness. We show a largely greater similarity within monozygotic than dizygotic pairs, resulting in a heritability estimate of 96%, not influenced by sleep need and intensity. If replicated, these results will establish the electroencephalographic profile during sleep as one of the most heritable traits of humans.

Synaptic learning rules and sparse coding in a model sensory system.

Neural circuits exploit numerous strategies for encoding information. Although the functional significance of individual coding mechanisms has been investigated, ways in which multiple mechanisms interact and integrate are not well understood. The locust olfactory system, in which dense, transiently synchronized spike trains across ensembles of antenna lobe (AL) neurons are transformed into a sparse representation in the mushroom body (MB; a region associated with memory), provides a well-studied preparation for investigating the interaction of multiple coding mechanisms. Recordings made in vivo from the insect MB demonstrated highly specific responses to odors in Kenyon cells (KCs). Typically, only a few KCs from the recorded population of neurons responded reliably when a specific odor was presented. Different odors induced responses in different KCs. Here, we explored with a biologically plausible model the possibility that a form of plasticity may control and tune synaptic weights of inputs to the mushroom body to ensure the specificity of KCs' responses to familiar or meaningful odors. We found that plasticity at the synapses between the AL and the MB efficiently regulated the delicate tuning necessary to selectively filter the intense AL oscillatory output and condense it to a sparse representation in the MB. Activity-dependent plasticity drove the observed specificity, reliability, and expected persistence of odor representations, suggesting a role for plasticity in information processing and making a testable prediction about synaptic plasticity at AL-MB synapses.

Functional EEG topography in sleep and waking: state-dependent and state-independent features.

Power spectra in the non-rapid eye movement sleep (NREMS) electroencephalogram (EEG) have been shown to exhibit frequency-specific topographic features that may point to functional differences in brain regions. Here, we extend the analysis to rapid eye movement sleep (REMS) and waking (W) to determine the extent to which EEG topography is determined by state under two different levels of sleep pressure. Multichannel EEG recordings were obtained from young men during a baseline night, a 40-h waking period, and a recovery night. Sleep deprivation enhanced EEG power in the low-frequency range (1-8 Hz) in all three vigilance states. In NREMS, the effect was largest in the delta band, in W, in the theta band, while in REMS, there was a peak in both the delta and the theta band. The response of REMS to prolonged waking and its pattern of EEG topography was intermediate between NREMS and W. Cluster analysis revealed a major topographic segregation into three frequency bands (1-8 Hz, 9-15 Hz, 16-24 Hz), which was largely independent of state and sleep pressure. To assess individual topographic traits within each state, the differences between pairs of power maps were compared within (i.e., for baseline and recovery) and between individuals (i.e., separately for baseline and recovery). A high degree of intraindividual correspondence of the power maps was observed. The frequency-specific clustering of power maps suggests that distinct generators underlie EEG frequency bands. Although EEG power is modulated by state and sleep pressure, basic topographic features appear to be state-independent.