Labeling of Good_ID for clusinfo

[ryanlash2]

Correct me if I'm mistaken on my interpretation of the code if BC good ids are ultimately the ones that are used. If we have a pipeline setup to use bombcell for quality assessment, would it mess up the pipeline to use cluster_bc_unitType.tsv for determining Good_ID that are used in clusinfo instead of the cluster_group.tsv?

[EnnyvanBeest]

There are multiple versions (matlab/Python), so if you can add a link to the specific line of code you are referring to, it will help me answering your question better.
In general you should be able to 'overwrite' clusinfo.Good_ID however you want though, so if you'd like to use your own method, you can just overwrite clusinfo.Good_ID after it has been initialized.

[ryanlash2]

Yes it is from the most recent version in ExtractKilosortData: `%% Load Cluster Info
myClusFile = dir(fullfile(KiloSortPaths{subsesid}, 'cluster_info.tsv')); % If you did phy (manual curation) we will find this one... We can trust you, right?
if isempty(myClusFile) || Params.RunQualityMetrics
if ~Params.RunQualityMetrics
disp('This data is not curated with phy! Hopefully you''re using automated quality metrics to find good units!')
end
curratedflag = 0;
myClusFile = dir(fullfile(KiloSortPaths{subsesid}, 'cluster_group.tsv'));
try
if isempty(myClusFile)
clusinfo = tdfread(fullfile(KiloSortPaths{subsesid}, 'cluster_KSLabel.tsv'));
else
clusinfo = tdfread(fullfile(myClusFile(1).folder, myClusFile(1).name));
end

    catch ME
        clusinfo = tdfread(fullfile(KiloSortPaths{subsesid}, 'cluster_KSLabel.tsv'));
        keyboard
    end

    % Convert sp data to correct cluster according to phy (clu and
    % template are not necessarily the same after splitting/merging)
    try
        [clusinfo, sp, emptyclus] = RemovingEmptyClusters(clusinfo, sp);
    catch ME
       keyboard
    end
    
    %clusidtmp = clusinfo.cluster_id;
    sp.clu = int32(sp.clu);
    % clusinfo.cluster_id = unique(sp.spikeTemplates);
    clusidtmp = int32(clusinfo.cluster_id);
    tmpLabel = char(length(clusinfo.cluster_id));
    KSLabelfile = tdfread(fullfile(KiloSortPaths{subsesid}, 'cluster_KSLabel.tsv'));
    tmpLabel(ismember(clusinfo.cluster_id, KSLabelfile.cluster_id)) = KSLabelfile.KSLabel(ismember(KSLabelfile.cluster_id, clusinfo.cluster_id));
    Label = [Label, tmpLabel];
    totSpkNum = histc(sp.clu, clusinfo.cluster_id);
    Good_IDtmp = ismember(tmpLabel, 'g') & totSpkNum' > paramBC.minNumSpikes; %Identify good clusters

    % Find depth and channel
    depthtmp = nan(length(clusinfo.cluster_id), 1);
    xtmp = nan(length(clusinfo.cluster_id), 1);
    channeltmp = nan(length(clusinfo.cluster_id), 1);
    for clusid = 1:length(depthtmp)
        % Depth according to PyKS2 output
        depthtmp(clusid) = round(sp.templateDepths(clusid)); %round(nanmean(sp.spikeDepths(find(sp.clu==clusid-1))));
        xtmp(clusid) = sp.templateXpos(clusid);
        [~, minidx] = min(cell2mat(arrayfun(@(X) pdist(cat(1, channelpostmp(X, :), [xtmp(clusid), depthtmp(clusid)]), 'euclidean'), 1:size(channelpostmp, 1), 'UniformOutput', 0)));
        try
            channeltmp(clusid) = channelmaptmp(minidx);
            depthtmp(clusid) = channelpostmp(minidx, 2);
            xtmp(clusid) = channelpostmp(minidx, 1);
        catch
            channeltmp(clusid) = channelmaptmp(minidx-1);
            depthtmp(clusid) = channelpostmp(minidx-1, 2);
            xtmp(clusid) = channelpostmp(minidx-1, 1);
        end
        sp.spikeDepths(ismember(sp.clu, clusidtmp(clusid))) = depthtmp(clusid);

    end
    depth = cat(1, depth, depthtmp);
    channel = cat(1, channel, channeltmp);

else
    disp('You did manual curation. You champion. If you have not enough time, maybe consider some automated algorithm...')
    CurationDone = 1;
    save(fullfile(KiloSortPaths{subsesid}, 'CuratedResults.mat'), 'CurationDone')
    myClusFile = dir(fullfile(KiloSortPaths{subsesid}, 'cluster_group.tsv'));
    clusinfo = tdfread(fullfile(myClusFile(1).folder, myClusFile(1).name));
    if ~isfield(clusinfo,'cluster_id')
        clusinfo.cluster_id = clusinfo.id;
    end
    % Convert sp data to correct cluster according to phy (clu and
    % template are not necessarily the same after  splitting/merging)
    [clusinfo, sp, emptyclus] = RemovingEmptyClusters(clusinfo, sp);
    if ~any(~isnan(clusinfo.group))
        disp('clusinfo.group is empty, taking KS labels')
        clusinfo.group = clusinfo.KSLabel;
    end
    curratedflag = 1;
    if isfield(clusinfo, 'cluster_id')
        clusidtmp = clusinfo.cluster_id;
        cluster_id = [cluster_id, clusinfo.cluster_id];
    else
        keyboard
        disp('Someone thought it was nice to change the name again...')
    end


    Label = [Label, clusinfo.group];`

[EnnyvanBeest]

In this part of the script, if there's a 'cluster_group.tsv' file available in the kilosort directory it will use that. If that's not available, it will try 'cluster_KSLabel.tsv' instead.
Later in the same script (line 635) Good_IDtmp will be overwritten by curated findings from bombcell if they are available, so if you run bombcell before or as part of unitmatch (see from line 479 in the same script) you should be fine.

But as mentioned, you can always do your own preprocessing pipeline and/or overwrite clusinfo.Good_ID later, if you are not sure this specific preprocessing pipeline works for you.