User Tools

Site Tools


analysis:course-w16:week16

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
analysis:course-w16:week16 [2016/02/28 21:31]
mvdm
analysis:course-w16:week16 [2016/03/01 09:50]
eirvine [Steps 5-7: Get co-activation probabilities]
Line 1: Line 1:
 ~~DISCUSSION~~ ~~DISCUSSION~~
- 
-:!: **UNDER CONSTRUCTION,​ PLEASE DO NOT USE YET** :!: 
  
 ===== Pairwise co-occurrence ===== ===== Pairwise co-occurrence =====
Line 8: Line 6:
  
   * Learn to think about possible ensemble firing patterns and ways to characterize them   * Learn to think about possible ensemble firing patterns and ways to characterize them
 +  * Implement, in detail, an analysis of pairwise co-occurrence during putative "​replay"​ events
   * Apply a permutation test (shuffle) to determine levels of chance (independent) co-occurrence   * Apply a permutation test (shuffle) to determine levels of chance (independent) co-occurrence
  
Line 34: Line 33:
 ==== Co-activation:​ concept and overall workflow ====  ==== Co-activation:​ concept and overall workflow ==== 
  
-Co-activation or co-occurrence analysis allows us to examine the content of "​replay"​ without direct reference to the field order of place cells in the environment. By grouping place cells into categories such as "​left-arm place cells" and "​right-arm place cells" we can ask if left cells are significantly more active together than right cells, which would suggest that the rat may have been recalling the left arm more frequently than the right arm.+Co-activation or co-occurrence analysis allows us to examine the content of "​replay"​ without direct reference to the field order of place cells in the environment ​(Cheng and Frank, 2008). By grouping place cells into categories such as "​left-arm place cells" and "​right-arm place cells" we can ask if left cells are significantly more active together than right cells, which would suggest that the rat may have been recalling the left arm more frequently than the right arm.
  
 {{ :​analysis:​course-w16:​cc_schematic1.png?​nolink&​300 |}} {{ :​analysis:​course-w16:​cc_schematic1.png?​nolink&​300 |}}
Line 50: Line 49:
   - **Get z-scored coactivity for cell pairs**: are the cell pairs co-active at greater than chance levels? (If the cells are firing randomly, we still expect them to co-occur to some degree)   - **Get z-scored coactivity for cell pairs**: are the cell pairs co-active at greater than chance levels? (If the cells are firing randomly, we still expect them to co-occur to some degree)
  
 +This module will first take you through how existing code accomplishes these steps. Then, the final section implements a basic version of the same analysis from scratch, so that you can get to know the guts of the analysis and build on it yourself.
 ==== Step-by-step ==== ==== Step-by-step ====
  
 === Setup and data loading === === Setup and data loading ===
  
-First, make sure you do a ''​git pull''​ as usual, and get the data. We'll be using session ''​R064-2015-04-22''​. You'll also need to include the ''​tasks\Alyssa-Tmaze''​ and ''​tasks\ReplayAnalysis''​ folders in your path.+First, make sure you do a ''​git pull''​ as usual, and get the data. We'll be using session ''​R064-2015-04-22''​. You'll also need to include the ''​tasks\Alyssa-Tmaze''​ and ''​tasks\Replay_Analysis''​ folders in your path.
  
 Then, we load the data: Then, we load the data:
Line 98: Line 98:
 evt = TSDtoIV(cfg,​SWR);​ % evt is the set of candidate events evt = TSDtoIV(cfg,​SWR);​ % evt is the set of candidate events
  
-clearvars -except evt ExpKeys+clearvars -except ​S pos evt ExpKeys
 </​code>​ </​code>​
 +
 +:!: Make sure that the filter above performed as expected. In particular, if the %%FieldTrip%% version of ''​filtfilt()''​ takes precedence in your path over the matlab builtin version, this could fail!
  
 Normally there is a lot more that goes on during candidate detection, such as the elimination of events that occur when the rat is moving too quickly, and but we will keep it simple for now. Normally there is a lot more that goes on during candidate detection, such as the elimination of events that occur when the rat is moving too quickly, and but we will keep it simple for now.
Line 251: Line 253:
 </​code>​ </​code>​
  
 +As you can see from the command line output of ''​SelectTS()'',​ not many cells of the original 178 are left after this operation!
 === Step 4: Make a Q-matrix === === Step 4: Make a Q-matrix ===
  
Line 264: Line 267:
 </​code>​ </​code>​
  
-This Q-matrix contains the number of times 10 different ​left-arm place cells spiked during ​1017 candidate, or sharp wave-ripple,​ events.+This Q-matrix contains the number of times each uniquely ​left-arm place cell spiked during candidate, or sharp wave-ripple,​ events.
  
-Consider column ​17 for the "​left"​ cells (''​Q.L(17,:​)''​):​ cell spiked three times, cell 5 spiked ​twice, and cell 4 spiked once. These left arm cells co-occurred during the same candidate interval. This matrix is the basis for all subsequent analysis steps, much in the same way that a slightly different Q-matrix is the basis for the [[analysis:​course-w16:​week10|decoding analysis]] covered earlier.+Consider column ​7 (i.e. event 7) for the "​left"​ cells (''​Q.L.data(:,7)''​):​ cell spiked three twice, and cells 11 and 12 spiked once. These left arm cells co-occurred during the same candidate interval. This matrix is the basis for all subsequent analysis steps, much in the same way that a slightly different Q-matrix is the basis for the [[analysis:​course-w16:​week10|decoding analysis]] covered earlier.
  
 === Steps 5-7: Get co-activation probabilities === === Steps 5-7: Get co-activation probabilities ===
Line 342: Line 345:
 set(gca,'​XLim',​[location(1)-1 location(2)+1],'​XTick',​location,'​XTickLabel',​{'​L'​ '​R'​});​ xlabel('​Field location'​) set(gca,'​XLim',​[location(1)-1 location(2)+1],'​XTick',​location,'​XTickLabel',​{'​L'​ '​R'​});​ xlabel('​Field location'​)
 ylabel({'​SWR coactivation';​ '​Z-score'​}) ylabel({'​SWR coactivation';​ '​Z-score'​})
-title(Coactivation above chance levels (p4))+title('Coactivation above chance levels (p4)')
 </​code>​ </​code>​
  
Line 354: Line 357:
 ylabels = {{'​Proportion of';'​SWRs active'​},​{'​Cell pair'; 'joint probability'​},​{'​SWR coactivation';​ '​Z-score'​}};​ ylabels = {{'​Proportion of';'​SWRs active'​},​{'​Cell pair'; 'joint probability'​},​{'​SWR coactivation';​ '​Z-score'​}};​
 arms = {'​L','​R'​};​ arms = {'​L','​R'​};​
 +colors = flipud(linspecer(2));​
 location = [1 2.5]; location = [1 2.5];
  
Line 362: Line 366:
         ​         ​
     h(iP) = subplot(1,​3,​iP);​     h(iP) = subplot(1,​3,​iP);​
-    for iBar = 1:2+    for iBar = 1:length(arms)
     ​     ​
         bar(location(iBar),​p_data(iP,​iBar),'​FaceColor',​colors(iBar,:​),'​EdgeColor','​none'​)         bar(location(iBar),​p_data(iP,​iBar),'​FaceColor',​colors(iBar,:​),'​EdgeColor','​none'​)
Line 377: Line 381:
 </​code>​ </​code>​
  
-This gives:+This gives something like:
  
-(insert results here)+{{ :​analysis:​course-w16:​cooccur_results.png?​nolink&​600 |}}
  
-It looks like his right cell pairs are more co-active than his left cell pairs. ​Could this have something ​to do with how often he experienced the right vs left trajectory?+It looks like his right cell pairs are more co-active than his left cell pairs, relative to shuffled co-coccurrence,​ suggesting that there may be more right trajectories replayed compared to left trajectories. 
 + 
 +Note: in interpreting results like the above, it is important ​to consider what asymmetries may be present in the neural and behavioral data that could account for any differences. We used the function ''​GetMatchedTrials()''​ earlier to create a fair comparison between left and right from a behavioral sampling perspective,​ but that doesn'​t change the fact that behaviorally,​ the animal ​experienced the right trajectory ​more often than the left:
  
 <code matlab> <code matlab>
Line 388: Line 394:
 </​code>​ </​code>​
  
-Maybe? But let's not discuss that. INSTEAD let's talk about more exciting details relating to the internal workings of CoOccurQ, because that's clearly more interesting than discussions about the contents of rat thoughts and whether the government will someday be able to probe our minds.+==== Implementing a basic co-occurrence analysis from scratch ====
  
-=== Interpreting ​coactivation analysis ​results ===+In this section, we will implement a simple version of coactivation analysis, where we don't care how many times a cell spiked during a given candidate event, only whether it spiked or not. Since our Q matrix contains spike counts, we convert the Q matrix to binary with 1's indicating that a cell was active in a candidate bin and 0's indicating that it wasn'​t:​
  
 +<code matlab>
 +% Pull some Q data out of the collector struct. Let's work with just the
 +% left cells for these examples:
 +Q_binary = Q.L.data;
  
 +% First exclude counts > 0
 +Q_binary(Q_binary > 1) = 1;
 +</​code>​
 +
 +== Calculating p0 ==
 +
 +This is a one-liner:
 +
 +<code matlab>
 +% Get expected co-occurrence under independence assumption
 +p0 = nanmean(Q_binary,​2);​ % fraction of bins each cell participates in individually
 +
 +% this is the same as nansum(Q_binary,​2)./​length(evt.tstart)
 +% Fraction of bins active is equal to the sum of candidate bins the cell was
 +% active in divided by the total number of candidates
 +
 +% Plot results
 +figure; bar(1, nanmean(p0),'​facecolor',​colors(1,:​));​ set(gca,'​xlim',​[0 2],'​ylim',​get(gca,'​ylim'​)*1.3)
 +</​code>​
 +
 +== Calculating p3 ==
 +
 +<code matlab>
 +% Get observed co-occurrences
 +nCells = size(Q_binary,​1);​
 +p3 = nan(nCells);​
 +for iI = 1:nCells
 +    for iJ = 1:nCells
 +   
 +        p3(iI,iJ) = nanmean(Q_binary(iI,:​).*Q_binary(iJ,:​));​
 +        ​
 +    end
 +end
 +p3(logical(eye(size(p3)))) = NaN; % probability of cell co-occurring with itself is not defined
 +</​code>​
 +
 +p3 is a symmetrical matrix that is nCells x nCells in dimension. The co-occurrence of cells 1 and 2, for example, can be seen in (row 1, column 2 ) and (row 2, column 1).
 +Note the diagonal of NaNs: this is because it doesn’t make sense asking about a cell’s co-occurrence with itself, so we exclude these values from the set.
 +
 +Since p3 is symmetrical,​ it has multiple copies of the same measure. We can keep one copy by extracting either the upper or lower triangular parts of the matrix:
 +
 +<code matlab>
 +keep = triu(ones(nCells),​1);​ % get indices of values we want to keep (the 1 means we're discarding the diagonal)
 +p3_upper = p3(find(keep));​
 +% now p3 is a vector and contains only one copy of each co-occurrence value
 +
 +% Plot results
 +figure; bar(1, nanmean(p3_upper),'​facecolor',​colors(1,:​));​ set(gca,'​xlim',​[0 2],'​ylim',​get(gca,'​ylim'​)*1.3)
 +</​code>​
 +
 +== Calculating p4 ==
 +
 +It's possible that co-occurrence values in p3 are expected by chance alone: if one set of cells (left or right) is more chatty overall and active in a higher proportion of candidate events, we would expect them to co-occur more often. ​
 +What we can do to check if the co-occurrences are similar to chance is to randomly shuffle the values within each row, and compare the shuffled co-occurrences with the observed co-occurrences and get a z-score. This will tell us if observed co-occurrences (p3) occur at a higher rate than expected by chance. The number of shuffles for accurate estimates of chance co-occurrence is at least 10,000. But a lower number reduces the time required for computation and is a close enough approximation for these purposes:
 +
 +<code matlab>
 +nShuffles = 1000;
 +
 +shuf_p4 = nan(nShuffles,​nCells,​nCells);​
 +nBins = size(Q_binary,​2);​
 +
 +for iShuf = 1:nShuffles
 +    ​
 +    % permute Q-matrix
 +    Q_shuffle = Q_binary;
 +    for iCell = 1:nCells
 +        Q_shuffle(iCell,:​) = Q_shuffle(iCell,​randperm(nBins));​ % this step using randperm mixes the contents of each row horizontally
 +    end
 +    ​
 +    % now compute shuffled co-occurrences
 +    for iI = 1:nCells
 +        for iJ = 1:nCells
 +            ​
 +            shuf_p4(iShuf,​iI,​iJ) = nanmean(Q_shuffle(iI,:​).*Q_shuffle(iJ,:​));​ % shuf_p4 is the shuffled co-occurrences and is a 3-dimensional array
 +            ​
 +        end
 +    end % of loop over cell pairs
 +    ​
 +end % of loop over shuffles
 +
 +% z-score coactivity
 +p4 = nan(nCells);​
 +
 +for iI = 1:nCells
 +    for iJ = 1:nCells
 +        ​
 +        p4(iI,iJ) = (p3(iI,​iJ)-nanmean(sq(shuf_p4(:,​iI,​iJ))))./​nanstd(sq(shuf_p4(:,​iI,​iJ)));​
 +        ​
 +    end
 +end % of loop over cell pairs
 +
 +% p4 is a symmetrical matrix like p3, so let's keep half of it:
 +p4_upper = p4(find(keep));​
 +figure; bar(1, nanmean(p4_upper),'​facecolor',​colors(1,:​));​ set(gca,'​xlim',​[0 2],'​ylim',​get(gca,'​ylim'​)*1.3)
 +</​code>​
 +
 +This shuffling procedure is an example of **resampling**:​ we create a number of different data sets based on some rearrangement of the original data. In this case, the specific resampling is a shuffle or permutation,​ which breaks any relationship between neurons (because we shuffle each neuron independently) and therefore functions as a control for the amount of co-occurrence we expect by chance (i.e. if the neurons were independently active). In general, a major advantage of resampling methods is that they preserve aspects of the underlying distribution -- in this case, of spike counts -- and make no particular assumptions about its shape, whereas many parametric statistical tests require data to be e.g. normally distributed.
  
 ==== Challenges ==== ==== Challenges ====
  
 ★ Why don't we just look at single-cell activation? Under what conditions would the results from that be the same, or different, from pairwise co-occurrence?​ ★ Why don't we just look at single-cell activation? Under what conditions would the results from that be the same, or different, from pairwise co-occurrence?​
 +
 +★ Comment on the choice of bins used in constructing the Q-matrix. It it reasonable to assume that SWR events are always 100ms in length? Modify the code to use the actual length of SWR events. Is the resampling statistic still doing the right thing in this case?
 +
 +★ Implement co-occurrence analysis on your own data.
  
 ==== Credits ==== ==== Credits ====
  
 This module was developed by [[https://​github.com/​aacarey | Alyssa Carey]]. This module was developed by [[https://​github.com/​aacarey | Alyssa Carey]].
analysis/course-w16/week16.txt · Last modified: 2018/07/07 10:19 (external edit)