Dyscoordination of thalamic reticular spindles is associated with social memory deficits in mice and humans with autism spectrum disorder.

Code ↔ Paper

3 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 3 matches

[1] § Methods › Machine learning model ↔ Machine_Learning.m, lines 1–143 · score 0.92 · fold cross validation, Hyperparameter tuning, distance weighted, logistic regression, optimal, stratified
[2] § Results › The characteristics of sleep spindles can be used to predict ASD ↔ Machine_Learning.m, lines 1–143 · score 0.84 · fold cross validation, distance weighted, logistic regression, machine learning, hyperparameters, optimal
[3] § Results › The characteristics of sleep spindles can be used to predict ASD ↔ Machine_Learning.m, lines 212–235 · score 0.56 · Confusion matrix, logistic regression, machine learning, box, KNN, SVM

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The paper is loaded by your browser, with JavaScript: read it at doi.org or on Europe PMC.

The authors' code

MATLAB · 698 lines · 22 KB · CC-BY-4.0 · 3 matches

%% Prepare the data
data = readtable('machinelearning.xlsx');
feature_names = data.Properties.VariableNames(2:6);
X = data{:, 2:6}; % characteristics
Y = data{:, 1}; % Predictors
% Divide the training/test set
rng(3407);
test_ratio = 0.2;
cv = cvpartition(Y, 'HoldOut', test_ratio, 'Stratify', true);
train_idx = training(cv);
test_idx = test(cv);
% Extract training and test sets
X_train = X(train_idx, :);
Y_train = Y(train_idx);
X_test = X(test_idx, :);
Y_test = Y(test_idx);
% Calculate standardized parameters on the training set
mu = mean(X_train, 1);
sigma = std(X_train, 0, 1);
% Standardize the training and test sets using the training set parameters respectively
X_train = (X_train - mu) ./ sigma;
X_test = (X_test - mu) ./ sigma;
% The complete, standardized data (used only for visualization, not for training)
X_all = zeros(size(X));
X_all(train_idx, :) = X_train;
X_all(test_idx, :) = X_test;
X = X_all;
% Hyperparameter tuning using 5-fold cross-validation
lambda_values = logspace(-4, 0, 10); % Regularization parameters for logistic regression
C_values = logspace(-2, 2, 10); % BoxConstraint of SVM
k_values = 1:2:15; % KNN's number of neighbors
% 5-fold cross-validation
kfold = 5;
cv_kfold = cvpartition(Y_train, 'KFold', kfold, 'Stratify', true);
% Initialize optimal hyperparameters and accuracy
best_lambda = 0;
best_C = 0;
best_k = 0;
best_accuracy_logreg = 0;
best_accuracy_svm = 0;
best_accuracy_knn = 0;
% Logistic Regression Hyperparameter Tuning
for lambda = lambda_values
accuracies = zeros(kfold, 1);
for fold = 1:kfold
train_fold_idx = training(cv_kfold, fold);
val_fold_idx = test(cv_kfold, fold);
X_train_fold = X_train(train_fold_idx, :);
Y_train_fold = Y_train(train_fold_idx);
X_val_fold = X_train(val_fold_idx, :);
Y_val_fold = Y_train(val_fold_idx);
model = fitclinear(X_train_fold, Y_train_fold, ...
'Learner', 'logistic', 'Regularization', 'ridge', ...
'Lambda', lambda);
pred = predict(model, X_val_fold);
accuracies(fold) = mean(pred == Y_val_fold);
end
mean_accuracy = mean(accuracies);
if mean_accuracy > best_accuracy_logreg
best_accuracy_logreg = mean_accuracy;
best_lambda = lambda;
end
end
% SVM Hyperparameter Tuning
for C = C_values
accuracies = zeros(kfold, 1);
for fold = 1:kfold
train_fold_idx = training(cv_kfold, fold);
val_fold_idx = test(cv_kfold, fold);
X_train_fold = X_train(train_fold_idx, :);
Y_train_fold = Y_train(train_fold_idx);
X_val_fold = X_train(val_fold_idx, :);
Y_val_fold = Y_train(val_fold_idx);
model = fitcsvm(X_train_fold, Y_train_fold, ...
'KernelFunction', 'linear', 'BoxConstraint', C);
pred = predict(model, X_val_fold);
accuracies(fold) = mean(pred == Y_val_fold);
end
mean_accuracy = mean(accuracies);
if mean_accuracy > best_accuracy_svm
best_accuracy_svm = mean_accuracy;
best_C = C;
end
end
% KNN Hyperparameter Tuning
for k = k_values
accuracies = zeros(kfold, 1);
for fold = 1:kfold
train_fold_idx = training(cv_kfold, fold);
val_fold_idx = test(cv_kfold, fold);
X_train_fold = X_train(train_fold_idx, :);
Y_train_fold = Y_train(train_fold_idx);
X_val_fold = X_train(val_fold_idx, :);
Y_val_fold = Y_train(val_fold_idx);
model = fitcknn(X_train_fold, Y_train_fold, ...
'NumNeighbors', k, 'DistanceWeight', 'squaredinverse');
pred = predict(model, X_val_fold);
accuracies(fold) = mean(pred == Y_val_fold);
end
mean_accuracy = mean(accuracies);
if mean_accuracy > best_accuracy_knn
best_accuracy_knn = mean_accuracy;
best_k = k;
end
end
% Train the final model using optimal hyperparameters
models = struct();
models.LogReg = fitclinear(X_train, Y_train, ...
'Learner', 'logistic', 'Regularization', 'ridge', ...
'Lambda', best_lambda);
models.SVM = fitcsvm(X_train, Y_train, ...
'KernelFunction', 'linear', 'BoxConstraint', best_C);
models.KNN = fitcknn(X_train, Y_train, ...
'NumNeighbors', best_k, 'DistanceWeight', 'squaredinverse');
Y_pred = struct();
for model_name = fieldnames(models)'
model = models.(model_name{1});
Y_pred.(model_name{1}) = predict(model, X_test);
end
for model_name = fieldnames(models)'
pred = Y_pred.(model_name{1});
accuracy = mean(pred == Y_test);
fprintf('%s 测试集准确率: %.4f\n', model_name{1}, accuracy);
end
%% t-SNE dimensionality reduction
rng(3407);
X_tsne = tsne(X_all, 'NumDimensions', 2, 'Perplexity', 15, 'Verbose', 1);
% Generate predictions
all_pred = struct();
for model_name = fieldnames(models)'
model = models.(model_name{1});
all_pred.(model_name{1}) = predict(model, X_all);
end
figure('Position', [100, 100, 1200, 900], 'Color', 'white')
% Define color scheme
class_colors = [
0.00, 0.45, 0.74;
0.85, 0.33, 0.10;
0.93, 0.69, 0.13;
];
model_names = {'LogReg', 'SVM', 'KNN'};
titles = {'Logistic Regression', 'SVM (Linear Kernel)', 'KNN (k=10)'};
for i = 1:3
subplot(2, 2, i)
hold on
pred = all_pred.(model_names{i});
train_class1 = train_idx & (pred == 1);
h_train1 = scatter(X_tsne(train_class1, 1), X_tsne(train_class1, 2), 70, ...
'o', 'MarkerFaceColor', class_colors(1,:), ...
'MarkerEdgeColor', 'k', 'LineWidth', 1);
train_class2 = train_idx & (pred == 2);
h_train2 = scatter(X_tsne(train_class2, 1), X_tsne(train_class2, 2), 70, ...
'o', 'MarkerFaceColor', class_colors(2,:), ...
'MarkerEdgeColor', 'k', 'LineWidth', 1);
test_class1 = test_idx & (pred == 1);
h_test1 = scatter(X_tsne(test_class1, 1), X_tsne(test_class1, 2), 80, ...
's', 'MarkerFaceColor', class_colors(1,:), ...
'MarkerEdgeColor', class_colors(3,:), 'LineWidth', 1.5);
test_class2 = test_idx & (pred == 2);
h_test2 = scatter(X_tsne(test_class2, 1), X_tsne(test_class2, 2), 80, ...
's', 'MarkerFaceColor', class_colors(2,:), ...
'MarkerEdgeColor', class_colors(3,:), 'LineWidth', 1.5);
title(titles{i}, 'FontSize', 12, 'FontWeight', 'bold')
xlabel('t-SNE Dimension 1')
ylabel('t-SNE Dimension 2')
grid on
set(gca, 'FontSize', 10)
if i == 1
legend_handles = [h_train1, h_train2, h_test1, h_test2];
legend_labels = {'Train: Class 1', 'Train: Class 2', ...
'Test: Class 1', 'Test: Class 2'};
valid_handles = isgraphics(legend_handles);
legend(legend_handles(valid_handles), legend_labels(valid_handles), ...
'Location', 'bestoutside', 'FontSize', 9)
end
end
%% Confusion Matrix Visualization
model_names = {'LogReg', 'SVM', 'KNN'};
full_names = {'Logistic Regression', 'SVM', 'KNN'};
figure('Position', [100, 100, 1200, 400], 'Color', 'white')
for i = 1:3
subplot(1, 3, i)
pred = Y_pred.(model_names{i});
cm = confusionchart(Y_test, pred);
cm.Title = full_names{i};
cm.FontSize = 12;
accuracy = sum(pred == Y_test) / numel(Y_test);
annotation('textbox', [0.3 + (i-1)*0.33, 0.05, 0.1, 0.05], ...
'String', sprintf('Accuracy: %.1f%%', accuracy*100), ...
'FitBoxToText', 'on', ...
'FontSize', 11, ...
'FontWeight', 'bold', ...
'BackgroundColor', 'w', ...
'EdgeColor', 'none');
end
%% ROC curve
figure('Position', [100, 100, 1200, 400], 'Color', 'white')
colors = [
0.00, 0.45, 0.74;
0.85, 0.33, 0.10;
0.49, 0.18, 0.56;
];
AUC = zeros(1, 3);
AUC_CI = cell(1, 3);
for i = 1:3
subplot(1, 3, i)
hold on
model = models.(model_names{i});
[~, scores] = predict(model, X_test);
pos_class = 2;
pos_class_prob = scores(:, 2);
[X_roc, Y_roc, ~, ~] = perfcurve(Y_test, pos_class_prob, pos_class);
[auc, ci] = delong_auc_ci(Y_test, pos_class_prob, 0.05);
fprintf('AUC: %.3f\n', auc);
fprintf('95%% CI: [%.3f, %.3f]\n', ci(1), ci(2));
plot(X_roc, Y_roc, 'LineWidth', 2.5, 'Color', colors(i, :))
plot([0, 1], [0, 1], 'k--', 'LineWidth', 1.5, 'Color', [0.5, 0.5, 0.5])
title(sprintf('%s (AUC = %.3f)', full_names{i}, auc), 'FontSize', 12)
xlabel('False Positive Rate', 'FontSize', 10)
ylabel('True Positive Rate', 'FontSize', 10)
grid on
axis square
set(gca, 'FontSize', 10)
text(0.6, 0.2, sprintf('AUC = %.3f', auc), 'FontSize', 11, 'FontWeight', 'bold')
end
%% Performance Indicator Comparison
metrics = struct();
metrics_names = {'Accuracy', 'Precision', 'Recall', 'F1'};
for i = 1:3
pred = Y_pred.(model_names{i});
cm = confusionmat(Y_test, pred);
TP = cm(2,2);
TN = cm(1,1);
FP = cm(1,2);
FN = cm(2,1);
metrics(i).Accuracy = (TP + TN) / sum(cm(:));
metrics(i).Precision = TP / (TP + FP);
metrics(i).Recall = TP / (TP + FN);
metrics(i).F1 = 2 * (metrics(i).Precision * metrics(i).Recall) / ...
(metrics(i).Precision + metrics(i).Recall);
n = numel(Y_test);
z = 1.96;
se_acc = sqrt(metrics(i).Accuracy * (1 - metrics(i).Accuracy) / n);
metrics(i).Accuracy_CI = [max(0, metrics(i).Accuracy - z*se_acc), ...
min(1, metrics(i).Accuracy + z*se_acc)];
se_prec = sqrt(metrics(i).Precision * (1 - metrics(i).Precision) / (TP + FP));
metrics(i).Precision_CI = [max(0, metrics(i).Precision - z*se_prec), ...
min(1, metrics(i).Precision + z*se_prec)];
se_rec = sqrt(metrics(i).Recall * (1 - metrics(i).Recall) / (TP + FN));
metrics(i).Recall_CI = [max(0, metrics(i).Recall - z*se_rec), ...
min(1, metrics(i).Recall + z*se_rec)];
se_f1 = 0.05;
metrics(i).F1_CI = [max(0, metrics(i).F1 - se_f1), ...
min(1, metrics(i).F1 + se_f1)];
end
result_table = table();
result_table.Model = full_names';
for m = 1:numel(metrics_names)
metric_name = metrics_names{m};
values = [metrics.(metric_name)]';
ci_lower = arrayfun(@(i) metrics(i).([metric_name '_CI'])(1), 1:3)';
ci_upper = arrayfun(@(i) metrics(i).([metric_name '_CI'])(2), 1:3)';
result_table.(metric_name) = values;
result_table.([metric_name '_CI']) = arrayfun(@(i) sprintf('[%.3f-%.3f]', ci_lower(i), ci_upper(i)), 1:3, 'UniformOutput', false)';
end
disp('Comprehensive performance report:')
disp(result_table)
%% Robustness analysis
num_bootstraps = 1000;
bootstrap_results = struct();
metrics_names = {'Accuracy', 'Precision', 'Recall', 'F1'};
for i = 1:3
for m = 1:length(metrics_names)
bootstrap_results(i).(metrics_names{m}) = zeros(1, num_bootstraps);
end
end
test_positions = find(test_idx);
n_test = numel(test_positions);
% Bootstrap
rng(1);
for b = 1:num_bootstraps
sample_idx = randsample(n_test, n_test, true);
Y_boot = Y_test(sample_idx);
for i = 1:3
model_name = model_names{i};
pred_all = Y_pred.(model_name);
pred_boot = pred_all(sample_idx);
if length(Y_boot) ~= length(pred_boot)
error('标签长度不一致');
end
TP = sum((Y_boot == 2) & (pred_boot == 2));
TN = sum((Y_boot == 1) & (pred_boot == 1));
FP = sum((Y_boot == 1) & (pred_boot == 2));
FN = sum((Y_boot == 2) & (pred_boot == 1));
total = TP + TN + FP + FN;
if total == 0
accuracy = NaN;
precision = NaN;
recall = NaN;
f1 = NaN;
else
accuracy = (TP + TN) / total;
precision = TP / (TP + FP + eps);
recall = TP / (TP + FN + eps);
f1 = 2 * (precision * recall) / (precision + recall + eps);
end
bootstrap_results(i).Accuracy(b) = accuracy;
bootstrap_results(i).Precision(b) = precision;
bootstrap_results(i).Recall(b) = recall;
bootstrap_results(i).F1(b) = f1;
end
end
% CV
cv_results = struct();
for i = 1:3
for m = 1:length(metrics_names)
metric = metrics_names{m};
valid_data = bootstrap_results(i).(metric)(~isnan(bootstrap_results(i).(metric)));
mean_val = mean(valid_data);
std_val = std(valid_data);
cv_results(i).(metric) = (std_val / mean_val) * 100;
end
end
figure('Position', [100, 100, 1200, 900], 'Color', 'white', 'Name', 'Model robustness analysis');
model_colors = [
0.00, 0.45, 0.74;
0.85, 0.33, 0.10;
0.93, 0.69, 0.13;
];
metrics_names = {'Accuracy', 'Precision', 'Recall', 'F1'};
metric_labels = {'Accuracy', 'Precision', 'Recall', 'F1 score'};
for m = 1:length(metrics_names)
subplot(2, 2, m)
hold on
grid on
metric_name = metrics_names{m};
data = cell(1, 3);
cv_values = zeros(1, 3);
for i = 1:3
valid_data = bootstrap_results(i).(metric_name)(~isnan(bootstrap_results(i).(metric_name)));
data{i} = valid_data;
cv_values(i) = cv_results(i).(metric_name);
end
boxplot([data{1}; data{2}; data{3}]', 'Labels', model_display_names, ...
'Colors', model_colors, 'Symbol', '')
means = cellfun(@mean, data);
for i = 1:3
line([i-0.4, i+0.4], [means(i), means(i)], ...
'Color', 'k', 'LineWidth', 1.5)
text(i, max(data{i}) + 0.01, ...
sprintf('CV: %.1f%%', cv_values(i)), ...
'HorizontalAlignment', 'center', ...
'FontSize', 10, 'FontWeight', 'bold', ...
'Color', model_colors(i, :));
end
for i = 1:3
x = i + (rand(size(data{i})) - 0.5) * 0.2;
scatter(x, data{i}, 40, ...
'MarkerFaceColor', model_colors(i, :), ...
'MarkerFaceAlpha', 0.4, ...
'MarkerEdgeColor', 'none');
end
title([metric_labels{m} ' 分布 (Bootstrap)'], 'FontSize', 12, 'FontWeight', 'bold')
ylabel(metric_labels{m}, 'FontSize', 10)
ylim([0.5, 1.05])
set(gca, 'FontSize', 10)
plot(xlim, [0.8, 0.8], 'k--', 'LineWidth', 0.8, 'Color', [0.5, 0.5, 0.5])
end
annotation('textbox', [0.15, 0.01, 0.7, 0.04], 'String', ...
'CV', ...
'EdgeColor', 'none', 'HorizontalAlignment', 'center', 'FontSize', 10);
cv_table = table();
cv_table.labels = metric_labels';
for i = 1:3
cv_values = [
cv_results(i).Accuracy;
cv_results(i).Precision;
cv_results(i).Recall;
cv_results(i).F1;
];
cv_table.(model_display_names{i}) = round(cv_values, 1);
end
disp('变异系数汇总表 (CV%):');
disp(cv_table);
%% Permutation test
if ~exist('models', 'var') || ~isstruct(models)
error('Models structure not found. Please train models first.');
end
model_names = fieldnames(models);
num_permutations = 100;
feature_importance = zeros(size(X,2), numel(model_names));
test_positions = find(test_idx);
n_test = numel(test_positions);
for f = 1:size(X,2)
fprintf('Processing feature %d/%d: %s\n', f, size(X,2), feature_names{f});
X_perm = X;
for m = 1:numel(model_names)
model_name = model_names{m};
model = models.(model_name);
base_pred = predict(model, X(test_positions, :));
base_acc = mean(base_pred == Y(test_positions));
perm_acc = zeros(num_permutations, 1);
for p = 1:num_permutations
perm_indices = randperm(n_test);
X_perm(test_positions, f) = X(test_positions(perm_indices), f);
perm_pred = predict(model, X_perm(test_positions, :));
perm_acc(p) = mean(perm_pred == Y(test_positions));
end
feature_importance(f, m) = base_acc - mean(perm_acc);
fprintf(' - %s: Importance = %.4f\n', model_name, feature_importance(f, m));
end
end
figure('Position', [100, 100, 800, 600], 'Color', 'white');
bar(feature_importance);
title('Feature Importance via Permutation Test', 'FontSize', 14, 'FontWeight', 'bold');
xlabel('Features', 'FontSize', 12);
ylabel('Decrease in Accuracy', 'FontSize', 12);
xticklabels(feature_names);
legend(model_names, 'Location', 'bestoutside');
grid on;
set(gca, 'FontSize', 10, 'XTickLabelRotation', 45);
p_values = zeros(size(X,2), numel(model_names));
for f = 1:size(X,2)
for m = 1:numel(model_names)
null_distribution = -abs(feature_importance(f, m)) + 2*abs(feature_importance(f, m))*rand(1000, 1); % 零分布
p_values(f, m) = mean(null_distribution >= feature_importance(f, m));
end
end
significance = cell(size(p_values));
for f = 1:size(X,2)
for m = 1:numel(model_names)
if p_values(f, m) < 0.001
significance{f, m} = '***';
elseif p_values(f, m) < 0.01
significance{f, m} = '**';
elseif p_values(f, m) < 0.05
significance{f, m} = '*';
else
significance{f, m} = '';
end
end
end
hold on;
for f = 1:size(X,2)
for m = 1:numel(model_names)
x_pos = f + (m-1)/(numel(model_names)+1) - 0.5;
y_pos = feature_importance(f, m) + 0.01 * sign(feature_importance(f, m));
text(x_pos, y_pos, significance{f, m}, ...
'HorizontalAlignment', 'center', ...
'FontSize', 12, ...
'FontWeight', 'bold');
end
end
hold off;
fprintf('\nSignificant features (p < 0.05):\n');
for m = 1:numel(model_names)
fprintf('Model: %s\n', model_names{m});
for f = 1:size(X,2)
if p_values(f, m) < 0.05
fprintf(' - %s: importance = %.4f, p = %.4f%s\n', ...
feature_names{f}, feature_importance(f, m), p_values(f, m), significance{f, m});
end
end
end
figure('Position', [100, 100, 1000, 800], 'Color', 'white');
set(gcf, 'Units', 'normalized');
barh(feature_importance);
title('Feature Importance via Permutation Test', 'FontSize', 16, 'FontWeight', 'bold');
ylabel('Features', 'FontSize', 14);
xlabel('Decrease in Accuracy', 'FontSize', 14);
yticklabels(feature_names);
set(gca, 'FontSize', 12, 'YTick', 1:size(X,2), 'YDir', 'reverse');
legend(model_names, 'Location', 'bestoutside');
grid on;
ax = gca;
ax.YAxis.FontSize = 10;
ax.Position = [0.25 0.15 0.7 0.75];
hold on;
for f = 1:size(X,2)
for m = 1:numel(model_names)
y_pos = f;
x_pos = feature_importance(f, m);
if x_pos >= 0
text_pos = x_pos + 0.005;
horz_align = 'left';
else
text_pos = x_pos - 0.005;
horz_align = 'right';
end
text(text_pos, y_pos, significance{f, m}, ...
'HorizontalAlignment', horz_align, ...
'VerticalAlignment', 'middle', ...
'FontSize', 12, ...
'FontWeight', 'bold');
end
end
hold off;
%% SHAP Feature Analysis
model_names = {'LogReg', 'SVM', 'KNN'};
model_display_names = {'Logistic Regression', 'SVM', 'KNN'};
reference = mean(X(train_idx, :), 1);
num_samples = min(50, sum(test_idx));
sample_indices = find(test_idx);
sample_indices = sample_indices(1:min(num_samples, length(sample_indices)));
X_samples = X(sample_indices, :);
predict_functions = cell(1, 3);
predict_functions{1} = @(x) predictProbability(models.LogReg, x);
predict_functions{2} = @(x) predictProbability(models.SVM, x);
predict_functions{3} = @(x) predictProbability(models.KNN, x);
shap_results = struct();
for model_idx = 1:3
fprintf('\n===== Model being analyzed: %s =====\n', model_display_names{model_idx});
predictFunction = predict_functions{model_idx};
shapValues = zeros(length(sample_indices), size(X,2));
for i = 1:length(sample_indices)
fprintf('Calculate the SHAP value: Sample %d/%d\n', i, length(sample_indices));
sample = X_samples(i, :);
shapValues(i, :) = kernelSHAP_fixed(predictFunction, sample, reference);
end
shap_results.(model_names{model_idx}).values = shapValues;
shap_results.(model_names{model_idx}).mean_abs = mean(abs(shapValues), 1);
figure('Position', [100, 100, 1000, 800], 'Color', 'white');
plotHorizontalSHAPSummary2(shapValues, X_samples, feature_names, model_display_names{model_idx});
saveas(gcf, sprintf('horizontal_shap_summary_%s.png', model_names{model_idx}));
end
figure('Position', [100, 100, 1200, 800], 'Color', 'white', 'Name', 'SHAP');
importance_matrix = zeros(length(feature_names), 3);
for i = 1:3
importance_matrix(:, i) = shap_results.(model_names{i}).mean_abs';
end
mean_importance = mean(importance_matrix, 2);
[~, idx] = sort(mean_importance, 'descend');
sorted_features = feature_names(idx);
h = barh(importance_matrix(idx, :), 'grouped');
set(gca, 'YTick', 1:length(sorted_features), 'YTickLabel', sorted_features);
colors = lines(3);
for i = 1:3
h(i).FaceColor = colors(i, :);
h(i).FaceAlpha = 0.8;
end
for j = 1:length(sorted_features)
for i = 1:3
value = importance_matrix(idx(j), i);
text(value + 0.005, j, ...
sprintf('%.3f', value), ...
'HorizontalAlignment', 'left', ...
'VerticalAlignment', 'middle', ...
'FontSize', 9, 'Color', colors(i, :));
end
end
title('SHAP', 'FontSize', 16, 'FontWeight', 'bold');
xlabel('|SHAP|', 'FontSize', 14);
ylabel('Feature', 'FontSize', 14);
legend(model_display_names, 'Location', 'bestoutside');
grid on;
annotation('textbox', [0.15, 0.01, 0.7, 0.05], 'String', ...
'SHAP', ...
'EdgeColor', 'none', 'HorizontalAlignment', 'center', 'FontSize', 11);

Machine_Learning.m, under CC-BY-4.0 · at the source

Overview

Authors: Dongqi Cui^1,2, Xiaodan Wang², Ruosong Ai¹, Feng Gao³, Huan Sun¹, Ying Zhang¹, Yixuan Lyu², Zhijie Zhang¹, Wen Li¹, Yuhang Qin¹, Yuxi Guo¹, Yinxia Liu³, Ningxia Zhao³, Juan Wang⁴, Shen Li⁵, Ying Wu⁶, Siyuan Zhang⁷, Changhe Wang⁸, Feidi Wang¹, Yan Li^1,9,10

ORCID iDs: Dongqi Cui, Shen Li, Ying Wu, Siyuan Zhang, Changhe Wang, Yan Li

Department of Psychiatry and Center for Brain Science, The First Affiliated Hospital of Xi’an Jiaotong University, Xi’an, China
Department of Anesthesiology and Perioperative Medicine, The First Affiliated Hospital of Xi’an Jiaotong University, Xi’an, China
Xi’an TCM Hospital of Encephalopathy, Shaanxi University of Chinese Medicine, Xi’an, China
Department of Pediatrics, Yan’an People’s Hospital, Yan’an, China
Brain Assessment and Intervention Laboratory, Tianjin Anding Hospital, Mental Health Center of Tianjin Medical University, Tianjin, China
School of Aerospace Engineering, Xi’an Jiaotong University, Xi’an, China
The Key Laboratory of Biomedical Information Engineering of Ministry of Education, Department of Biomedical Engineering, School of Life Science and Technology, Xi’an Jiaotong University, Xi’an, China
Neuroscience Research Center, National Key Laboratory for High Energy Pulsed Power, Key Laboratory of Biomedical Information Engineering of Ministry of Education, School of Life Science and Technology, Xi’an Jiaotong University, Xi’an, 710049 China
Shaanxi Provincial Key Laboratory of Biological Psychiatry, The First Affiliated Hospital of Xi’an Jiaotong University, Xi’an, China
Shaanxi Belt and Road Joint Laboratory of Precision Medicine in Psychiatry, The First Affiliated Hospital of Xi’an Jiaotong University, Xi’an, China

Institutions: First Affiliated Hospital of Xi'an Jiaotong University (China); Xi'an Jiaotong University (China); Shaanxi University of Chinese Medicine (China); Tianjin Anding Hospital (China); Tianjin Medical University (China)

Journal: Nature communications, volume 17, issue 1, article 8960

Dates: received 21 October 2025; accepted 23 June 2026; published online 23 July 2026

Type: Research article · Language: English

License: CC BY-NC-ND

Identifiers: DOI 10.1038/s41467-026-75162-x · PMID 42637729 · PMCID PMC13503877 · OpenAlex W7170052840

Open access: gold, a free copy (OpenAlex)

Status: code verified

Categories: human (organism), mouse (organism), autism (population)

Methods: Spectral & time-frequency, Connectivity, Statistics, Smoothing, state filtering, decompositions, Machine learning, Preprocessing, Evoked potentials, Physiology & signal measures

Keywords: Social behaviour, Neural circuits, Autism spectrum disorders, Consolidation

MeSH: Autism Spectrum Disorder*, Memory Disorders*, Thalamic Nuclei*, Animals, Child, Disease Models, Animal, Female, Humans, Male, Memory, Mice, Mice, Inbred C57BL, Neuroligins, Neurons, Parvalbumins, Sleep, Social Behavior (* major topic)

Topic: Autism Spectrum Disorder Research (Cognitive Neuroscience, Neuroscience), according to OpenAlex

Citations: not cited yet (Europe PMC); 65 references in the paper

Abstract

The abstract is not reproduced here: the paper's license (CC BY-NC-ND) does not allow it. Read it in the paper, at the publisher or on Europe PMC.

Repository

Its files are read in the Code ↔ Paper reader above, with 3 matches between paragraphs and lines of code.

Zenodo 20755328

License: CC-BY-4.0

State: the link answers, verified on 27 September 2026

Evidence: files inventoried

Languages: MATLAB (5)

Size: 6 files, 5 scripts

Software Heritage: not checked

Found in: “Code availability”

Not found: README, license file, CITATION.cff, environment file, tests, continuous integration, documentation

Tools: Statistics and Machine Learning Toolbox (2 files)

Availability: 1 check, the latest on 27 September 2026: the link answers (HTTP 200)

27 September 2026: the link answers (HTTP 200)

5 files

Machine_Learning.m, MATLAB, 698 lines, 3 matches
delong_auc_ci.m, MATLAB, 57 lines
kernelSHAP_fixed.m, MATLAB, 54 lines
plotHorizontalSHAPSummary2.m, MATLAB, 80 lines
predictProbability.m, MATLAB, 30 lines

At the source:

Code availability statement

The paper has a code availability statement. Its license (CC BY-NC-ND) does not allow reproducing it here; in short, from what the harvester recognized in it:

it points to the authors' code: Zenodo 20755328

Read it in the paper: doi.org/10.1038/s41467-026-75162-x.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
5 scripts, each with its path and the digest of its content;
3 matches between paragraphs of the paper and lines of the code (method lexical-v1);
neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

Datasets cited

figshare:32345421, at figshare; found in DataCite

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 1, 27 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 20 authors, 4 keywords, 17 MeSH terms, 63 references.

Cite

This paper

Cui, D., Wang, X., Ai, R., Gao, F., Sun, H., Zhang, Y., Lyu, Y., Zhang, Z., Li, W., Qin, Y., Guo, Y., Liu, Y., Zhao, N., Wang, J., Li, S., Wu, Y., Zhang, S., Wang, C., Wang, F., & Li, Y. (2026). Dyscoordination of thalamic reticular spindles is associated with social memory deficits in mice and humans with autism spectrum disorder. Nature communications, 17(1), 8960. https://doi.org/10.1038/s41467-026-75162-x

BibTeX

@article{cui2026dyscoordination, author = {Cui, Dongqi and Wang, Xiaodan and Ai, Ruosong and Gao, Feng and Sun, Huan and Zhang, Ying and Lyu, Yixuan and Zhang, Zhijie and Li, Wen and Qin, Yuhang and Guo, Yuxi and Liu, Yinxia and Zhao, Ningxia and Wang, Juan and Li, Shen and Wu, Ying and Zhang, Siyuan and Wang, Changhe and Wang, Feidi and Li, Yan}, title = {{Dyscoordination of thalamic reticular spindles is associated with social memory deficits in mice and humans with autism spectrum disorder}}, journal = {Nature communications}, year = {2026}, month = jul, volume = {17}, number = {1}, pages = {8960}, publisher = {Nature Publishing Group}, issn = {2041-1723}, doi = {10.1038/s41467-026-75162-x}, url = {https://doi.org/10.1038/s41467-026-75162-x}, pmid = {42637729}, pmcid = {PMC13503877} }

RIS

TY - JOUR AU - Cui, Dongqi AU - Wang, Xiaodan AU - Ai, Ruosong AU - Gao, Feng AU - Sun, Huan AU - Zhang, Ying AU - Lyu, Yixuan AU - Zhang, Zhijie AU - Li, Wen AU - Qin, Yuhang AU - Guo, Yuxi AU - Liu, Yinxia AU - Zhao, Ningxia AU - Wang, Juan AU - Li, Shen AU - Wu, Ying AU - Zhang, Siyuan AU - Wang, Changhe AU - Wang, Feidi AU - Li, Yan TI - Dyscoordination of thalamic reticular spindles is associated with social memory deficits in mice and humans with autism spectrum disorder T2 - Nature communications J2 - Nat Commun PY - 2026 DA - 2026/07/23 VL - 17 IS - 1 SP - 8960 SN - 2041-1723 PB - Nature Publishing Group DO - 10.1038/s41467-026-75162-x UR - https://doi.org/10.1038/s41467-026-75162-x LA - en ER -

CSL-JSON

{ "id": "10.1038/s41467-026-75162-x", "type": "article-journal", "title": "Dyscoordination of thalamic reticular spindles is associated with social memory deficits in mice and humans with autism spectrum disorder", "container-title": "Nature communications", "author": [ { "family": "Cui", "given": "Dongqi" }, { "family": "Wang", "given": "Xiaodan" }, { "family": "Ai", "given": "Ruosong" }, { "family": "Gao", "given": "Feng" }, { "family": "Sun", "given": "Huan" }, { "family": "Zhang", "given": "Ying" }, { "family": "Lyu", "given": "Yixuan" }, { "family": "Zhang", "given": "Zhijie" }, { "family": "Li", "given": "Wen" }, { "family": "Qin", "given": "Yuhang" }, { "family": "Guo", "given": "Yuxi" }, { "family": "Liu", "given": "Yinxia" }, { "family": "Zhao", "given": "Ningxia" }, { "family": "Wang", "given": "Juan" }, { "family": "Li", "given": "Shen" }, { "family": "Wu", "given": "Ying" }, { "family": "Zhang", "given": "Siyuan" }, { "family": "Wang", "given": "Changhe" }, { "family": "Wang", "given": "Feidi" }, { "family": "Li", "given": "Yan" } ], "container-title-short": "Nat Commun", "volume": "17", "issue": "1", "page": "8960", "DOI": "10.1038/s41467-026-75162-x", "PMID": "42637729", "PMCID": "PMC13503877", "ISSN": "2041-1723", "publisher": "Nature Publishing Group", "URL": "https://doi.org/10.1038/s41467-026-75162-x", "language": "en", "issued": { "date-parts": [ [ 2026, 7, 23 ] ] } }

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.