Inferring brain-wide interactions using data-constrained recurrent neural network models.

1. %{
2. =================================
3. This project contains the raw data on which this paper is based:
5. J.Minxha, R.Adolphs, S.Fusi, A.N.Mamelak, U.Rutishauser. Flexible recruitment of memory-based choice representations by human medial-frontal cortex. Science, 2020 (in press).
7. Permanent URL for data: http://doi.org/10.17605/OSF.IO/U3KCP
9. Please see http://www.rutishauserlab.org/ for up-to-date contact information.
11. This work has been made possible by grants from the National Institutes of Health (NIMH R01MH110831; The NIH BRAIN initiative U01NS103792; the National Science Foundation CAREER BCS-1554105; and a Memory&Cognitive Disorders Award from the McKnight Endowment for the Neurosciences).
12. =================================
14. ********* Main Files *********
15. There are two main files, "main_sfc.m" which reproduces the spike-field
16. coherence results in Figure 5 and "main_spikes.m", which reproduces all the raster and PSTHs. The cell indices for the rasters plotted in the main
17. figures of the paper are included in the file. The purpose of these files
18. is to illustrate how to load and utilize the data rather than to reproduce
19. all the figures in the paper. Note that all the data needed to produce the
20. findings in the paper is included.
22. ********* Spiking data structures *********
23. The spiking folder, "spikes" contains three files, "ha.mat" (amygddala and
24. hippocampus), "mfc.mat" (preSMA and dACC), and "mfc_cell_annotations.mat"
25. which contains annotations for the MFC population (indeces for choice cells, weights for the MFC population as assinged by the choice decoder, etc.) The structures in "ha.mat" and "mfc.mat" have the following fields:
27. sessionID: unique identifier for the sessions (can be usd to index into
28. other data structures, ex. SFC)
29. behavior: contains all the relevant behavior from the session in which the
30. cell was recorded. It is itself a structure with the following
31. fields:
33. RT: response time (in seconds)
34. iscorrect: boolean indicating if trial was correct
35. nr_appearances: ranges from 0 to 7, indicates the number of
36. times this image has been presented before in
37. this session. This can be converted into a
38. boolian variable indicating new/old by
39. thresholding at 1. So any image with an index of
40. 1 or greater, is part of the "old" set of stimuli.
41. categories: can take 4 possible values, 1 = cars, 4 = fruits,
42. 5 = human face, 7 = monkey face
43. target: this is relevant only for the categorization trials,
44. indicates target image category. For example, if the
45. value for a trial is "5" that means that for that trial,
46. the subject has to respond with "yes" if the stimulus
47. shown was a human face. the target changed by blocks.
48. For memory blocks, the values are "NaN"
49. response: the subjects response, "1" = yes, "0" = no.
50. trial_type: what task was being performed, two columns which are
51. the complement of each other. Column one indexes all
52. categorization trials, column two all memory trials.
53. block_nr: block number of the trial, should be 1-8, with the
54. exception of one session where there was an extra block
55. for training purposes.
56. gt: ground truth for the response.
57. response_type: indicates the effector to be used, 2 = eye or
58. 1 = button press.
60. ts: time stamps organized into trials, provided either aligned to
61. stimulus onset or to the response.
62. binsize : indicates the size of the bin in milliseconds, used to compute
63. firing rate for PSTHs
64. stepSize: for internal use, indicates the stepsize of the moving window
65. used to estimate firing rate for PSTHs, expressed as a fraction
66. of the bin size.
67. cellinfo: row 1 = cluster number (internal use), row 2 = channel number
68. (internal use), row 3 = area code of the brain area this cell
69. was in.
70. is_control: boolean indicating if this session was a control session.
71. See Figure S6 of the paper.
73. ********* Spike-field coherence (SFC) data structures *********
74. The folder "sfc" contains two subdirectories called "baseline" and "stim".
75. These names refer to the time window within which the spike-field coherence
76. was measured. In the case of "baseline" it's [-1 0] seconds relative to
77. stimulus onset. For "stim" it's [0.2 1.2] seconds after stimulus onset.
78. All SFC analysis in the paper is done during the baseline, with the
79. exception of the contrast between true positive trials and false negative
80. trials, which was done after the stimulus onset. The "stim" folder contains
81. an additional file, called "session_raw_stats". For the SFC analysis after
82. stimulus onset, we only used sessions where the referencing was done
83. locally, i.e. within the same cluster of electrodes as the one the cell was
84. recorded from. The reason for this, is that ERPs (event related potentials)
85. are significantly diminished for locally referenced sessions. The
86. referencing (local vs. distal) information is given in
87. "session_raw_stats.mat".
89. session_raw_stats.mat (in stim subdirectory) contains two structures,
90. "stats_amy" and "stats_hippo" containing referencing information for
91. amygdala and hippocampus respectively. The fields are:
93. name: this is the session identifier
94. areas: the area the electrodes were in, will read 'uLA','uRA' or 'uLH',
95. 'uRH' for left and right amygdala/hippocampus.
96. is_local: a boolean indicating if the session was locally referenced.
99. tp_vs_fn.mat (in stim subdirectory) contains two structures, "sfc_mfc_amy"
100. and "sfc_mfc_hippo" the coherence and spike-triggered power for all mfc
101. cells with respect to amygdala and hippmpus LFP respectively. These
102. structures apply to different LFP recordings but the structure is the same,
103. with the following fields:
105. f: the frequencies sampled (logarithmically from 2 t0 125Hz)
106. cellinfo: first row = cluster number (internal use only), second row =
107. channel number (internal),
108. third row = identifier for the brain area where it was recorded
109. 4,8 = left and right preSMA
110. 2,6 = left and right dACC
111. sessionid: unique identifier for the recording session
112. coh: contains PPC data, a cell array that has the same length as the
113. number of MFC cells (767).
114. If one entry in the array is empty, it is because that cell did
115. not meet requirements to be included. The main exclusion
116. criteria is that there must be at least 25 spikes per condition.
117. Given that false negatives are rare, this reduces the number of
118. cells included in the analysis. If the cell did meet the
119. requirements, then there will be N subarrays, where N is the
120. number of conditions +1. So for the TN vs. FN contrast, there
121. will be 3 subarrays at each location, where the first condition
122. is the union of the spikes used in TN and FN (the "all spikes"
123. condition).
124. stp : same as "coh" but contains the spike triggered power instead of
125. the PPC. The spike triggered power is estimated (like the
126. coherence) from the complex values returned by the fourier
127. transform of the around the spikes.
128. params: contains information on how the analysis was performed.
129. Contains subfields like "min_nr_spikes" which is the main
130. criteria used to determine if a cell is included in teh analysis.
132. In the "baseline" subdirectory, "sfc_data_task" contains two structures
133. that have exactly the same layout as what is described above for
134. "tp_vs_fn.mat"
137. ********* Area codes for brain regions *********
138. 1 = left amygdala
139. 2 = left dACC
140. 3 = left hippocampus
141. 4 = left preSMA
142. 5 = right amygdala
143. 6 = right dACC
144. 7 = right hippocampus
145. 8 = right preSMA
147. %}

README.m, no license · at the source