Neuroscience Cloud Analysis As a Service

Abstract

A major goal of computational neuroscience is to develop powerful analysis tools that operate on large datasets. These methods provide an essential toolset to unlock scientific insights from new experiments. Unfortunately, a major obstacle currently impedes progress: while existing analysis methods are frequently shared as open source software, the infrastructure needed to deploy these methods -at scale, reproducibly, cheaply, and quickly -remains totally inaccessible to all but a minority of expert users. As a result, many users can not fully exploit these tools, due to constrained computational resources (limited or costly compute hardware) and/or mismatches in expertise (experimentalists vs. large-scale computing experts). In this work we develop Neuroscience Cloud Analysis As a Service (NeuroCAAS): a fully-managed infrastructure platform, based on modern large-scale computing advances, that makes state-of-the-art data analysis tools accessible to the neuroscience community. We offer NeuroCAAS as an open source service with a drag-and-drop interface, entirely removing the burden of infrastructure expertise, purchasing, maintenance, and deployment. NeuroCAAS is enabled by three key contributions. First, NeuroCAAS cleanly separates tool implementation from usage, allowing cutting-edge methods to be served directly to the end user with no need to read or install any analysis software. Second, NeuroCAAS automatically scales as needed, providing reliable, highly elastic computational resources that are more efficient than personal or lab-supported hardware, without management overhead. Finally, we show that many popular data analysis tools offered through NeuroCAAS outperform typical analysis solutions (in terms of speed and cost) while improving ease of use and maintenance, dispelling the myth that cloud compute is prohibitively expensive and technically inaccessible. By removing barriers to fast, efficient cloud computation, NeuroCAAS can dramatically accelerate both the dissemination and the effective use of cutting-edge analysis tools for neuroscientific discovery.

A large-scale neural network training framework for generalized estimation of single-trial population dynamics

“…A key challenge with emerging, computationally-intensive data analysis methods is that the computational infrastructure and expertise necessary to make effective use of these tools is a significant barrier to widespread adoption (52). For example, many labs do not have the resources necessary to train dozens of models in parallel across many GPUs.…”

BARcode DEmixing through Non-negative Spatial Regression (BarDensr)

“…To develop an efficient implementation of BarDensr on the NeuroCAAS cloud platform (Abe et al, 2020), we needed to find the most cost-effective hardware for the job. Using a 1000 × 1000 sized image from the experimental data described in the main text, we ran the model on several different AWS instance types.…”

“…The method requires about two minutes of compute time on a p2.xlarge Amazon GPU instance to process a seven-round, four-channels 1000 × 1000-pixel field of view from an experiment targeting 79 different transcripts. We also provide an implementation for the NeuroCAAS web-service (Abe et al, 2020), which can be used in a drag-and-drop fashion, with no installation required. We compared this method with three alter- Figure 1: BarDensr uses non-negative regression to demix and deconvolve the observed image stack, yielding a sparse intensity image for each barcode.…”

BARcode DEmixing through Non-negative Spatial Regression (BarDensr)

“…The method requires about two minutes of compute time on a Amazon GPU instance to process a seven-round, four-channel 1000 × 1000-voxel field of view from an experiment targeting 79 different transcripts. We also provide an implementation for the NeuroCAAS web-service [ 12 ], which can be used in a drag-and-drop fashion, with no installation required. We compared this method with three alternatives: the ‘spot-based’ and ‘pixel-based’ methods of starfish ; a ‘blobs-first’ approach (Single Round Matching, or SRM, based on methods from [ 5 , 8 ]); and a ‘barcodes-first’ approach (Correlation approach, or ‘corr,’ based on [ 6 , 10 , 11 ]).…”