Inferring single-trial neural population dynamics using sequential auto-encoders

• Type of Material: Article
• Authors: Chethan Pandarinath, Emory University; Daniel J. O'Shea, Stanford University; Jasmine Collins, Google AI; Rafal Jozefowicz, Google AI; Sergey D. Stavisky, Stanford University; Jonathan C. Kao, Stanford University; Eric M. Trautmann, Stanford University; Matthew T. Kaufman, Stanford University; Stephen I. Ryu, Stanford University; Leigh R. Hochberg, Veteran Affairs Medical Center; Jaimie M. Henderson, Stanford University; Krishna V. Shenoy, Stanford University; L. F. Abbott, Columbia University; David Sussillo, Stanford University
• Language: English
• Date: 2018-10-01
• Publisher: Nature Research (part of Springer Nature)
• Publication Version: Author Accepted Manuscript (After Peer Review)
• Copyright Statement: © 2018, The Author(s), under exclusive licence to Springer Nature America, Inc.
• Final Published Version (URL): http://dx.doi.org/10.1038/s41592-018-0109-9
• Title of Journal or Parent Work: Nature Methods
• ISSN: 1548-7091
• Volume: 15
• Issue: 10
• Start Page: 805
• End Page: +
• Grant/Funding Information: L.R.H’s research was supported by NIH-NIDCD R01DC009899, Rehabilitation Research and Development Service, Department of Veterans Affairs (B6453R), MGH-Deane Institute for Integrated Research on Atrial Fibrillation and Stroke; Executive Committee on Research, Massachusetts General Hospital.; K.V.S. and J.M.H.’s research was supported by Stanford BioX-NeuroVentures, Stanford Institute for Neuro-Innovation and Translational Neuroscience, Garlick Foundation and Reeve Foundation.; S.D.S. was supported by the ALS Association’s Milton Safenowitz Postdoctoral Fellowship.; L.F.A.’s research was supported by US National Institutes of Health grant MH093338, the Gatsby Charitable Foundation through the Gatsby Initiative in Brain Circuitry at Columbia University, the Simons Foundation, the Swartz Foundation, the Harold and Leila Y. Mathers Foundation, and the Kavli Institute for Brain Science at Columbia University.; J.M.H.’s research was supported by NIH-NIDCD R01DC014034.; K.V.S.’s research was supported by the following awards: an NIH-NINDS award (T-R01NS076460), an NIH-NIMH award (T-R01MH09964703), an NIH Director’s Pioneer award (8DP1HD075623), a DARPA-DSO ‘REPAIR’ award (N66001–10-C-2010), a DARPA-BTO ‘NeuroFAST’ award (W911NF-14–2-0013), a Simons Foundation Collaboration on the Global Brain award (325380), and the Howard Hughes Medical Institute.; C.P. was supported by a postdoctoral fellowship from the Craig H. Neilsen Foundation for spinal cord injury research and the Stanford Dean’s Fellowship.
• Keywords: FIELD POTENTIAL OSCILLATIONS; GAUSSIAN-PROCESS; Science & Technology; CORTEX; MOVEMENTS; Life Sciences & Biomedicine; DEEP BRAIN-STIMULATION; Biochemical Research Methods; Biochemistry & Molecular Biology; PROSTHESIS
• Research Categories: Health Sciences, Medicine and Surgery; Biology, Neuroscience; Biology, Molecular

Abstract

Neuroscience is experiencing a revolution in which simultaneous recording of thousands of neurons is revealing population dynamics that are not apparent from single-neuron responses. This structure is typically extracted from data averaged across many trials, but deeper understanding requires studying phenomena detected in single trials, which is challenging due to incomplete sampling of the neural population, trial-to-trial variability, and fluctuations in action potential timing. We introduce latent factor analysis via dynamical systems, a deep learning method to infer latent dynamics from single-trial neural spiking data. When applied to a variety of macaque and human motor cortical datasets, latent factor analysis via dynamical systems accurately predicts observed behavioral variables, extracts precise firing rate estimates of neural dynamics on single trials, infers perturbations to those dynamics that correlate with behavioral choices, and combines data from non-overlapping recording sessions spanning months to improve inference of underlying dynamics.