Inferring the function performed by a recurrent neural network, bioRxiv, 2019-04-05

AbstractA central goal in systems neuroscience is to understand the functions performed by neural circuits. Previous top-down models addressed this question by comparing the behaviour of an ideal model circuit, optimised to perform a given function, with neural recordings. However, this requires guessing in advance what function is being performed, which may not be possible for many neural systems. Here, we propose an alternative approach that uses recorded neural responses to directly infer the function performed by a neural network. We assume that the goal of the network can be expressed via a reward function, which describes how desirable each state of the network is for carrying out a given objective. This allows us to frame the problem of optimising each neuron’s responses by viewing neurons as agents in a reinforcement learning (RL) paradigm; likewise the problem of inferring the reward function from the observed dynamics can be treated using inverse RL. Our framework encompasses previous influential theories of neural coding, such as efficient coding and attractor network models, as special cases, given specific choices of reward function. Finally, we can use the reward function inferred from recorded neural responses to make testable predictions about how the network dynamics will adapt depending on contextual changes, such as cell death andor varying input statistics, so as to carry out the same underlying function with different constraints.

biorxiv neuroscience 0-100-users 2019

PIRATE A fast and scalable pangenomics toolbox for clustering diverged orthologues in bacteria, bioRxiv, 2019-04-05

AbstractCataloguing the distribution of genes within natural bacterial populations is essential for understanding evolutionary processes and the genetic basis of adaptation. Here we present a pangenomics toolbox, PIRATE (Pangenome Iterative Refinement And Threshold Evaluation), which identifies and classifies orthologous gene families in bacterial pangenomes over a wide range of sequence similarity thresholds. PIRATE builds upon recent scalable software developments to allow for the rapid interrogation of thousands of isolates. PIRATE clusters genes (or other annotated features) over a wide range of amino-acid or nucleotide identity thresholds and uses the clustering information to rapidly classify paralogous gene families into either putative fissionfusion events or gene duplications. Furthermore, PIRATE orders the pangenome using a directed graph, provides a measure of allelic variation and estimates sequence divergence for each gene family. We demonstrate that PIRATE scales linearly with both number of samples and computation resources, allowing for analysis of large genomic datasets, and compares favorably to other popular tools. PIRATE provides a robust framework for analysing bacterial pangenomes, from largely clonal to panmictic species.AvailabilityPIRATE is implemented in Perl and is freely available under an GNU GPL 3 open source license from <jatsext-link xmlnsxlink=httpwww.w3.org1999xlink ext-link-type=uri xlinkhref=httpsgithub.comSionBaylissPIRATE>httpsgithub.comSionBaylissPIRATE<jatsext-link>.<jatssec sec-type=supplementary-material>Supplementary InformationSupplementary data is available online.

biorxiv bioinformatics 100-200-users 2019

The ELIXIR Core Data Resources fundamental infrastructure for the life sciences, bioRxiv, 2019-04-05

AbstractMotivationLife science research in academia, industry, agriculture, and the health sector is critically dependent on free and open data resources. ELIXIR, the European Research Infrastructure for life sciences data, has undertaken the task of identifying the set of Core Data Resources within Europe that are of most fundamental importance to the life science community for the long-term preservation of biological data. Having defined the Core Data Resources, we explored characteristics of the usage, impact and sustainability of the set as a whole to assess the value and importance of these resources as an infrastructure, to understand sustainability to the infrastructure, and to demonstrate a model for assessing Core Data Resources worldwide.ResultsThe nineteen resources designated as Core Data Resources by ELIXIR together form a data infrastructure in Europe that is a subset of the wider worldwide open life sciences data infrastructure. These resources are of crucial importance to research throughout the world. We show that, from 2013 to 2017, data managed by the Core Data Resources tripled and usage doubled while staff numbers increased by only a sixth. Additionally, support for the Core Data Resources is precarious, with all resources together having assured funding for less than a third of current staff after only three years.Our findings demonstrate the importance of the ELIXIR Core Data Resources as repositories for research data and the knowledge generated from those data, while also demonstrating the precarious nature of the funding environment for this infrastructure. The ELIXIR Core Data Resources are part of a larger worldwide life sciences data resources ecosystem. Both within Europe and as part of the Global Biodata Coalition, ELIXIR will work for longer-term support for the worldwide life sciences data resource infrastructure and for the subset of that infrastructure that is the ELIXIR Core Data Resources.

biorxiv bioinformatics 0-100-users 2019

Training and inferring neural network function with multi-agent reinforcement learning, bioRxiv, 2019-04-05

AbstractA central goal in systems neuroscience is to understand the functions performed by neural circuits. Previous top-down models addressed this question by comparing the behaviour of an ideal model circuit, optimised to perform a given function, with neural recordings. However, this requires guessing in advance what function is being performed, which may not be possible for many neural systems. To address this, we propose a new framework for optimising a recurrent network using multi-agent reinforcement learning (RL). In this framework, a reward function quantifies how desirable each state of the network is for performing a given function. Each neuron is treated as an ‘agent’, which optimises its responses so as to drive the network towards rewarded states. Three applications follow from this. First, one can use multi-agent RL algorithms to optimise a recurrent neural network to perform diverse functions (e.g. efficient sensory coding or motor control). Second, one could use inverse RL to infer the function of a recorded neural network from data. Third, the theory predicts how neural networks should adapt their dynamics to maintain the same function when the external environment or network structure changes. This could lead to theoretical predictions about how neural network dynamics adapt to deal with cell death andor varying sensory stimulus statistics.

biorxiv neuroscience 0-100-users 2019

 

Created with the audiences framework by Jedidiah Carlson

Powered by Hugo