Firefly task: perturbation, gain and density analyses

MATLAB analysis code behind

Alefantis P., Lakshminarasimhan K., Avila E., Noel J.-P., Pitkow X., Angelaki D.E. (2022). Sensory evidence accumulation using optic flow in a naturalistic navigation task. The Journal of Neuroscience, 42(27). doi:10.1523/JNEUROSCI.2203-21.2022

Humans and macaques steer with a joystick through a virtual ground plane to a remembered "firefly" location, using only the optic flow their own movement generates. To test whether they integrate that flow over time (rather than replaying a learned motor plan), the paper uses three manipulations, and this repository holds the analysis for each:

Module	Manipulation	What the code does
perturbation/	Unpredictable optic-flow perturbations that push the subject off course	The two core methods: gen_sim_ptb simulates the "ignored the perturbation" counterfactual by adding the perturbation velocity profile to matched unperturbed trajectories and integrating to a stopping point (the control against which real endpoints are compared by ROC/AUC); get_kernel estimates the velocity response kernel to the perturbation by regressing speed traces on an amplitude-scaled boxcar basis
gain/	Joystick gain (1, 1.5, 2) changes the consequence of every action	Pools sessions by gain (con_gain); travel-time CDFs per gain (time_cdf_fun, time_cdf); log-log regression of travel time on distance and speed (LOGregress_fun)
density/	Optic-flow density changes how much evidence the flow carries	Pools sessions by ground-plane density and splits endpoints per density level (con_dens)
utils/		Euclidean endpoint error, cell concatenation

How it fits with the lab pipeline

Raw sessions are imported and pre-processed by the Angelaki lab's firefly-monkey pipeline (Kaushik Lakshminarasimhan), which produces one behaviours object per session (trials with events, prs, logical, continuous, and stats with final positions and trial-type indices). Everything here consumes those objects; nothing from that pipeline is duplicated in this repository. ComputeROCFirefly (ROC analysis) also comes from firefly-monkey and must be on the path.

Running

MATLAB R2020a or later, Statistics and Machine Learning Toolbox (regress), Image Processing Toolbox (medfilt3), and firefly-monkey on the path. Point the scripts at your data:

setenv('FIREFLY_DATA_ROOT', '/path/to/sessions')   % folders m44/, m51/, m53/ with session .mat files
addpath(genpath(pwd))
time_cdf                                            % gain: travel-time CDFs per gain, all animals
m = con_dens(fullfile(getenv('FIREFLY_DATA_ROOT'),'m51'));   % density: pooled endpoints per density
[kernel, ts] = get_kernel(Yresp, Amplitudes, Yphant);        % perturbation: response kernel

Behavioural data are not distributed here; see the data statement in the paper.

Licence

MIT (see LICENSE).