A Machine Learning Enhanced Mechanistic Simulation Framework for Functional Deficit Prediction in TBI

Resting state functional magnetic resonance imaging (rsfMRI), and the underlying brain networks identified with it, have recently appeared as a promising avenue for the evaluation of functional deficits without the need for active patient participation. We hypothesize here that such alteration can be inferred from tissue damage within the network. From an engineering perspective, the numerical prediction of tissue mechanical damage following an impact remains computationally expensive. To this end, we propose a numerical framework aimed at predicting resting state network disruption for an arbitrary head impact, as described by the head velocity, location and angle of impact, and impactor shape. The proposed method uses a library of precalculated cases leveraged by a machine learning layer for efficient and quick prediction. The accuracy of the machine learning layer is illustrated with a dummy fall case, where the machine learning prediction is shown to closely match the full simulation results. The resulting framework is finally tested against the rsfMRI data of nine TBI patients scanned within 24 h of injury, for which paramedical information was used to reconstruct in silico the accident. While more clinical data are required for full validation, this approach opens the door to (i) on-the-fly prediction of rsfMRI alterations, readily measurable on clinical premises from paramedical data, and (ii) reverse-engineered accident reconstruction through rsfMRI measurements.

Machine learning of stochastic gene network phenotypes

A recurrent challenge in biology is the development of predictive quantitative models because most molecular and cellular parameters have unknown values and realistic models are analytically intractable. While the dynamics of the system can be analyzed via computer simulations, substantial computational resources are often required given uncertain parameter values resulting in large numbers of parameter combinations, especially when realistic biological features are included. Simulation alone also often does not yield the kinds of intuitive insights from analytical solutions. Here we introduce a general framework combining stochastic/mechanistic simulation of reaction systems and machine learning of the simulation data to generate computationally efficient predictive models and interpretable parameter-phenotype maps. We applied our approach to investigate stochastic gene expression propagation in biological networks, which is a contemporary challenge in the quantitative modeling of single-cell heterogeneity. We found that accurate, predictive machine-learning models of stochastic simulation results can be constructed. Even in the simplest networks existing analytical schemes generated significantly less accurate predictions than our approach, which revealed interesting insights when applied to more complex circuits, including the extensive tunability of information propagation enabled by feedforward circuits and how even single negative feedbacks can utilize stochastic fluctuations to generate robust oscillations. Our approach is applicable beyond biology and opens up a new avenue for exploring complex dynamical systems.

Drivers of geographical patterns of North American language diversity

Although many hypotheses have been proposed to explain why humans speak so many languages and why languages are unevenly distributed across the globe, the factors that shape geographical patterns of cultural and linguistic diversity remain poorly understood. Prior research has tended to focus on identifying universal predictors of language diversity, without accounting for how local factors and multiple predictors interact. Here, we use a unique combination of path analysis, mechanistic simulation modelling, and geographically weighted regression to investigate the broadly described, but poorly understood, spatial pattern of language diversity in North America. We show that the ecological drivers of language diversity are not universal or entirely direct. The strongest associations imply a role for previously developed hypothesized drivers such as population density, resource diversity, and carrying capacity with group size limits. The predictive power of this web of factors varies over space from regions where our model predicts approximately 86% of the variation in diversity, to areas where less than 40% is explained.