Code for the analyses in 'Feature-selective responses in macaque visual cortex follow eye movements during natural vision.'

The associated data are in this DANDI repository (neural and behavioral data) and this OSF repository (images; TTL events used in one analysis, script_2c).

So long as a man’s eyes are open in the light, the act of seeing is involuntary; that is, he cannot then help mechanically seeing whatever objects are before him. Nevertheless, any one’s experience will teach him, that though he can take in an undiscriminating sweep of things at one glance, it is quite impossible for him, attentively, and completely, to examine any two things—however large or however small—at one and the same instant of time; never mind if they lie side by side and touch each other. But if you now come to separate these two objects, and surround each by a circle of profound darkness; then, in order to see one of them, in such a manner as to bring your mind to bear on it, the other will be utterly excluded from your contemporary consciousness.

—Herman Melville, Moby-Dick

Examples

The folder test_results contains analysis code and results for an example session; the corresponding data are provided in test_data for convenience. The results include:

• Jupyter notebooks (.ipynb), one per analysis, that contain code, output log, and example plots;
• HDF5 (.h5) files that contain numerical results.

The subfolders are for analyses with alternative parameters (e.g., using a 200 ms response window instead of sliding 50 ms windows).

The results are produced by the bash script, run_one_sess_all_main.sh. The script provides both a demo and further documentation of analysis parameters and dependencies.

Repo structure

1. Analysis: Parameterized scripts (.ipynb) for per-session analyses.

2. Postprocessing: Scripts to extract data-estimated parameters used in downstream analyses. See Workflow overview for details.

3. lib: Code for frequently used subroutines.

4. db: Parameters for analyses and plotting, including image region labels (masks and bounding boxes), lists of neurons (all, visually responsive, and visually selective), and per-neuron parameters (latencies and RFs; more details below).

5. Image_features: Scripts to pre-calculate DNN image embeddings used in model-based analyses (Figs. 5 & 6).

6. Summary: Scripts for population-level statistics and summary; plots reproduce the paper main figures.

To reproduce the analyses

The script run_one_sess_all_main.sh can be applied to all sessions. The following notes further explain the pipeline and setup.

Workflow overview

Each session is analyzed independently*. For each analysis and session, the same script is run using the session name as a parameter, This is done in batch using papermill. The results are saved in .h5 files for use in summary plots and, sometimes, as inputs to downstream analyses.

*Some analyses use per-neuron parameters (latencies and RFs) estimated from data across sessions. This procedure is detailed in the paper Methods and the script run_one_sess_hier_group.sh. The folder db contains these parameters pre-calculated to facilitate reproducibility.

The code in Image_features is run once to pre-calculate image DNN embeddings (used in scripts 6a, 6b, 7a). This facilitates analyzing hundreds of sessions in parallel without needing GPUs. (Note that to cache the features needs ~200 GB of disk space for the default settings in the paper.)

File locations

See lib/local_paths.[py,json] for details on where to put files (preprocessed neural data and images) and where outputs will be saved by default. The analysis scripts assume all preprocessed data are in [data_root]/Preprocessed (no subfolders) and named like the files in test_data. Files downloaded from DANDI will have a different naming/folder-structure convention and need to be reorganized.

Requirements

Reasonably recent versions of Python and the following libraries. Generally used libraries include numpy, h5py, pandas, scipy, scikit-learn, numba, xarray, matplotlib, seaborn, and tqdm. Specific analyses and/or preprocessing use Pillow (PIL), scikit-image, opencv-python, and pytorch. We use papermill to batch-run (and log the results of) analyses over many sessions.