MELLODDY: Cross-pharma Federated Learning at Unprecedented Scale Unlocks Benefits in QSAR without Compromising Proprietary Information.

Federated multipartner machine learning has been touted as an appealing and efficient method to increase the effective training data volume and thereby the predictivity of models, particularly when the generation of training data is resource-intensive. In the landmark MELLODDY project, indeed, each of ten pharmaceutical companies realized aggregated improvements on its own classification or regression models through federated learning. To this end, they leveraged a novel implementation extending multitask learning across partners, on a platform audited for privacy and security. The experiments involved an unprecedented cross-pharma data set of 2.6+ billion confidential experimental activity data points, documenting 21+ million physical small molecules and 40+ thousand assays in on-target and secondary pharmacodynamics and pharmacokinetics. Appropriate complementary metrics were developed to evaluate the predictive performance in the federated setting. In addition to predictive performance increases in labeled space, the results point toward an extended applicability domain in federated learning. Increases in collective training data volume, including by means of auxiliary data resulting from single concentration high-throughput and imaging assays, continued to boost predictive performance, albeit with a saturating return. Markedly higher improvements were observed for the pharmacokinetics and safety panel assay-based task subsets.

Optimising sounds for the driving of sleep oscillations by closed-loop auditory stimulation.

Recent studies have shown that slow oscillations (SOs) can be driven by rhythmic auditory stimulation, which deepens slow-wave sleep (SWS) and improves memory and the immune-supportive hormonal milieu related to this sleep stage. While different attempts have been made to optimise the driving of the SOs by changing the number of click stimulations, no study has yet investigated the impact of applying more than five clicks in a row. Likewise, the importance of the type of sounds in eliciting brain responses is presently unclear. In a study of 12 healthy young participants (10 females; aged 18-26 years), we applied an established closed-loop stimulation method, which delivered sequences of 10 pink noises, 10 pure sounds (B note of 247 Hz), 10 pronounced "a" vowels, 10 sham, 10 variable sounds, and 10 "oddball" sounds on the up phase of the endogenous SOs. By analysing area under the curve, amplitude, and event related potentials, we explored whether the nature of the sound had a differential effect on driving SOs. We showed that every stimulus in a 10-click sequence, induces a SO response. Interestingly, all three types of sounds that we tested triggered SOs. However, pink noise elicited a more pronounced response compared to the other sounds, which was explained by a broader topographical recruitment of brain areas. Our data further suggest that varying the sounds may partially counteract habituation.

Dendritic sodium spikes endow neurons with inverse firing rate response to correlated synaptic activity.

Many neurons possess dendrites enriched with sodium channels and are capable of generating action potentials. However, the role of dendritic sodium spikes remain unclear. Here, we study computational models of neurons to investigate the functional effects of dendritic spikes. In agreement with previous studies, we found that point neurons or neurons with passive dendrites increase their somatic firing rate in response to the correlation of synaptic bombardment for a wide range of input conditions, i.e. input firing rates, synaptic conductances, or refractory periods. However, neurons with active dendrites show the opposite behavior: for a wide range of conditions the firing rate decreases as a function of correlation. We found this property in three types of models of dendritic excitability: a Hodgkin-Huxley model of dendritic spikes, a model with integrate and fire dendrites, and a discrete-state dendritic model. We conclude that fast dendritic spikes confer much broader computational properties to neurons, sometimes opposite to that of point neurons.

Slow feature analysis with spiking neurons and its application to audio stimuli.

Extracting invariant features in an unsupervised manner is crucial to perform complex computation such as object recognition, analyzing music or understanding speech. While various algorithms have been proposed to perform such a task, Slow Feature Analysis (SFA) uses time as a means of detecting those invariants, extracting the slowly time-varying components in the input signals. In this work, we address the question of how such an algorithm can be implemented by neurons, and apply it in the context of audio stimuli. We propose a projected gradient implementation of SFA that can be adapted to a Hebbian like learning rule dealing with biologically plausible neuron models. Furthermore, we show that a Spike-Timing Dependent Plasticity learning rule, shaped as a smoothed second derivative, implements SFA for spiking neurons. The theory is supported by numerical simulations, and to illustrate a simple use of SFA, we have applied it to auditory signals. We show that a single SFA neuron can learn to extract the tempo in sound recordings.

Ideomotor feedback control in a recurrent neural network.

The architecture of a neural network controlling an unknown environment is presented. It is based on a randomly connected recurrent neural network from which both perception and action are simultaneously read and fed back. There are two concurrent learning rules implementing a sort of ideomotor control: (i) perception is learned along the principle that the network should predict reliably its incoming stimuli; (ii) action is learned along the principle that the prediction of the network should match a target time series. The coherent behavior of the neural network in its environment is a consequence of the interaction between the two principles. Numerical simulations show a promising performance of the approach, which can be turned into a local and better "biologically plausible" algorithm.

A biological gradient descent for prediction through a combination of STDP and homeostatic plasticity.

Identifying, formalizing, and combining biological mechanisms that implement known brain functions, such as prediction, is a main aspect of research in theoretical neuroscience. In this letter, the mechanisms of spike-timing-dependent plasticity and homeostatic plasticity, combined in an original mathematical formalism, are shown to shape recurrent neural networks into predictors. Following a rigorous mathematical treatment, we prove that they implement the online gradient descent of a distance between the network activity and its stimuli. The convergence to an equilibrium, where the network can spontaneously reproduce or predict its stimuli, does not suffer from bifurcation issues usually encountered in learning in recurrent neural networks.

Federated learning for predicting histological response to neoadjuvant chemotherapy in triple-negative breast cancer.

Triple-negative breast cancer (TNBC) is a rare cancer, characterized by high metastatic potential and poor prognosis, and has limited treatment options. The current standard of care in nonmetastatic settings is neoadjuvant chemotherapy (NACT), but treatment efficacy varies substantially across patients. This heterogeneity is still poorly understood, partly due to the paucity of curated TNBC data. Here we investigate the use of machine learning (ML) leveraging whole-slide images and clinical information to predict, at diagnosis, the histological response to NACT for early TNBC women patients. To overcome the biases of small-scale studies while respecting data privacy, we conducted a multicentric TNBC study using federated learning, in which patient data remain secured behind hospitals' firewalls. We show that local ML models relying on whole-slide images can predict response to NACT but that collaborative training of ML models further improves performance, on par with the best current approaches in which ML models are trained using time-consuming expert annotations. Our ML model is interpretable and is sensitive to specific histological patterns. This proof of concept study, in which federated learning is applied to real-world datasets, paves the way for future biomarker discovery using unprecedentedly large datasets.

Coupling radiative, conductive and convective heat-transfers in a single Monte Carlo algorithm: A general theoretical framework for linear situations.

It was recently shown that radiation, conduction and convection can be combined within a single Monte Carlo algorithm and that such an algorithm immediately benefits from state-of-the-art computer-graphics advances when dealing with complex geometries. The theoretical foundations that make this coupling possible are fully exposed for the first time, supporting the intuitive pictures of continuous thermal paths that run through the different physics at work. First, the theoretical frameworks of propagators and Green's functions are used to demonstrate that a coupled model involving different physical phenomena can be probabilized. Second, they are extended and made operational using the Feynman-Kac theory and stochastic processes. Finally, the theoretical framework is supported by a new proposal for an approximation of coupled Brownian trajectories compatible with the algorithmic design required by ray-tracing acceleration techniques in highly refined geometry.

Hebbian learning of recurrent connections: a geometrical perspective.

We show how a Hopfield network with modifiable recurrent connections undergoing slow Hebbian learning can extract the underlying geometry of an input space. First, we use a slow and fast analysis to derive an averaged system whose dynamics derives from an energy function and therefore always converges to equilibrium points. The equilibria reflect the correlation structure of the inputs, a global object extracted through local recurrent interactions only. Second, we use numerical methods to illustrate how learning extracts the hidden geometrical structure of the inputs. Indeed, multidimensional scaling methods make it possible to project the final connectivity matrix onto a Euclidean distance matrix in a high-dimensional space, with the neurons labeled by spatial position within this space. The resulting network structure turns out to be roughly convolutional. The residual of the projection defines the nonconvolutional part of the connectivity, which is minimized in the process. Finally, we show how restricting the dimension of the space where the neurons live gives rise to patterns similar to cortical maps. We motivate this using an energy efficiency argument based on wire length minimization. Finally, we show how this approach leads to the emergence of ocular dominance or orientation columns in primary visual cortex via the self-organization of recurrent rather than feedforward connections. In addition, we establish that the nonconvolutional (or long-range) connectivity is patchy and is co-aligned in the case of orientation learning.

The future of digital health with federated learning.

Data-driven machine learning (ML) has emerged as a promising approach for building accurate and robust statistical models from medical data, which is collected in huge volumes by modern healthcare systems. Existing medical data is not fully exploited by ML primarily because it sits in data silos and privacy concerns restrict access to this data. However, without access to sufficient data, ML will be prevented from reaching its full potential and, ultimately, from making the transition from research to clinical practice. This paper considers key factors contributing to this issue, explores how federated learning (FL) may provide a solution for the future of digital health and highlights the challenges and considerations that need to be addressed.

A Deep Learning Architecture for Temporal Sleep Stage Classification Using Multivariate and Multimodal Time Series.

Sleep stage classification constitutes an important preliminary exam in the diagnosis of sleep disorders. It is traditionally performed by a sleep expert who assigns to each 30 s of the signal of a sleep stage, based on the visual inspection of signals such as electroencephalograms (EEGs), electrooculograms (EOGs), electrocardiograms, and electromyograms (EMGs). We introduce here the first deep learning approach for sleep stage classification that learns end-to-end without computing spectrograms or extracting handcrafted features, that exploits all multivariate and multimodal polysomnography (PSG) signals (EEG, EMG, and EOG), and that can exploit the temporal context of each 30-s window of data. For each modality, the first layer learns linear spatial filters that exploit the array of sensors to increase the signal-to-noise ratio, and the last layer feeds the learnt representation to a softmax classifier. Our model is compared to alternative automatic approaches based on convolutional networks or decisions trees. Results obtained on 61 publicly available PSG records with up to 20 EEG channels demonstrate that our network architecture yields the state-of-the-art performance. Our study reveals a number of insights on the spatiotemporal distribution of the signal of interest: a good tradeoff for optimal classification performance measured with balanced accuracy is to use 6 EEG with 2 EOG (left and right) and 3 EMG chin channels. Also exploiting 1 min of data before and after each data segment offers the strongest improvement when a limited number of channels are available. As sleep experts, our system exploits the multivariate and multimodal nature of PSG signals in order to deliver the state-of-the-art classification performance with a small computational cost.

Performance of an Ambulatory Dry-EEG Device for Auditory Closed-Loop Stimulation of Sleep Slow Oscillations in the Home Environment.

Recent research has shown that auditory closed-loop stimulation can enhance sleep slow oscillations (SO) to improve N3 sleep quality and cognition. Previous studies have been conducted in lab environments. The present study aimed to validate and assess the performance of a novel ambulatory wireless dry-EEG device (WDD), for auditory closed-loop stimulation of SO during N3 sleep at home. The performance of the WDD to detect N3 sleep automatically and to send auditory closed-loop stimulation on SO were tested on 20 young healthy subjects who slept with both the WDD and a miniaturized polysomnography (part 1) in both stimulated and sham nights within a double blind, randomized and crossover design. The effects of auditory closed-loop stimulation on delta power increase were assessed after one and 10 nights of stimulation on an observational pilot study in the home environment including 90 middle-aged subjects (part 2).The first part, aimed at assessing the quality of the WDD as compared to a polysomnograph, showed that the sensitivity and specificity to automatically detect N3 sleep in real-time were 0.70 and 0.90, respectively. The stimulation accuracy of the SO ascending-phase targeting was 45 ± 52°. The second part of the study, conducted in the home environment, showed that the stimulation protocol induced an increase of 43.9% of delta power in the 4 s window following the first stimulation (including evoked potentials and SO entrainment effect). The increase of SO response to auditory stimulation remained at the same level after 10 consecutive nights. The WDD shows good performances to automatically detect in real-time N3 sleep and to send auditory closed-loop stimulation on SO accurately. These stimulation increased the SO amplitude during N3 sleep without any adaptation effect after 10 consecutive nights. This tool provides new perspectives to figure out novel sleep EEG biomarkers in longitudinal studies and can be interesting to conduct broad studies on the effects of auditory stimulation during sleep.

Addressing nonlinearities in Monte Carlo.

Monte Carlo is famous for accepting model extensions and model refinements up to infinite dimension. However, this powerful incremental design is based on a premise which has severely limited its application so far: a state-variable can only be recursively defined as a function of underlying state-variables if this function is linear. Here we show that this premise can be alleviated by projecting nonlinearities onto a polynomial basis and increasing the configuration space dimension. Considering phytoplankton growth in light-limited environments, radiative transfer in planetary atmospheres, electromagnetic scattering by particles, and concentrated solar power plant production, we prove the real-world usability of this advance in four test cases which were previously regarded as impracticable using Monte Carlo approaches. We also illustrate an outstanding feature of our method when applied to acute problems with interacting particles: handling rare events is now straightforward. Overall, our extension preserves the features that made the method popular: addressing nonlinearities does not compromise on model refinement or system complexity, and convergence rates remain independent of dimension.

A local Echo State Property through the largest Lyapunov exponent.

Echo State Networks are efficient time-series predictors, which highly depend on the value of the spectral radius of the reservoir connectivity matrix. Based on recent results on the mean field theory of driven random recurrent neural networks, enabling the computation of the largest Lyapunov exponent of an ESN, we develop a cheap algorithm to establish a local and operational version of the Echo State Property.

Regular graphs maximize the variability of random neural networks.

In this work we study the dynamics of systems composed of numerous interacting elements interconnected through a random weighted directed graph, such as models of random neural networks. We develop an original theoretical approach based on a combination of a classical mean-field theory originally developed in the context of dynamical spin-glass models, and the heterogeneous mean-field theory developed to study epidemic propagation on graphs. Our main result is that, surprisingly, increasing the variance of the in-degree distribution does not result in a more variable dynamical behavior, but on the contrary that the most variable behaviors are obtained in the regular graph setting. We further study how the dynamical complexity of the attractors is influenced by the statistical properties of the in-degree distribution.

Relative entropy minimizing noisy non-linear neural network to approximate stochastic processes.

A method is provided for designing and training noise-driven recurrent neural networks as models of stochastic processes. The method unifies and generalizes two known separate modeling approaches, Echo State Networks (ESN) and Linear Inverse Modeling (LIM), under the common principle of relative entropy minimization. The power of the new method is demonstrated on a stochastic approximation of the El Niño phenomenon studied in climate research.

Macroscopic equations governing noisy spiking neuronal populations with linear synapses.

Deriving tractable reduced equations of biological neural networks capturing the macroscopic dynamics of sub-populations of neurons has been a longstanding problem in computational neuroscience. In this paper, we propose a reduction of large-scale multi-population stochastic networks based on the mean-field theory. We derive, for a wide class of spiking neuron models, a system of differential equations of the type of the usual Wilson-Cowan systems describing the macroscopic activity of populations, under the assumption that synaptic integration is linear with random coefficients. Our reduction involves one unknown function, the effective non-linearity of the network of populations, which can be analytically determined in simple cases, and numerically computed in general. This function depends on the underlying properties of the cells, and in particular the noise level. Appropriate parameters and functions involved in the reduction are given for different models of neurons: McKean, Fitzhugh-Nagumo and Hodgkin-Huxley models. Simulations of the reduced model show a precise agreement with the macroscopic dynamics of the networks for the first two models.

Multiscale analysis of slow-fast neuronal learning models with noise.

This paper deals with the application of temporal averaging methods to recurrent networks of noisy neurons undergoing a slow and unsupervised modification of their connectivity matrix called learning. Three time-scales arise for these models: (i) the fast neuronal dynamics, (ii) the intermediate external input to the system, and (iii) the slow learning mechanisms. Based on this time-scale separation, we apply an extension of the mathematical theory of stochastic averaging with periodic forcing in order to derive a reduced deterministic model for the connectivity dynamics. We focus on a class of models where the activity is linear to understand the specificity of several learning rules (Hebbian, trace or anti-symmetric learning). In a weakly connected regime, we study the equilibrium connectivity which gathers the entire 'knowledge' of the network about the inputs. We develop an asymptotic method to approximate this equilibrium. We show that the symmetric part of the connectivity post-learning encodes the correlation structure of the inputs, whereas the anti-symmetric part corresponds to the cross correlation between the inputs and their time derivative. Moreover, the time-scales ratio appears as an important parameter revealing temporal correlations.