Speech clocks decode dementia phenotypes, social exposome, and biological aging.

Code ↔ Paper

10 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 10 matches

[1] § MATERIALS AND METHODS › Diagnostic criteria and clinical/cognitive assessments ↔ Code/linear_regressions.ipynb, lines 1445 to 1464 · score 0.66 · episodic memory, visual memory, global cognition, executive, AD, diagnostic · Report a mismatch
[2] § MATERIALS AND METHODS › Statistical analysis and predictor-based age estimation ↔ Code/regression_main.ipynb, lines 294 to 355 · score 0.64 · gradient boosting regressor, outer, nested, inner, R2, fold · Report a mismatch
[3] § RESULTS › SAGs and clinical/cognitive domains ↔ Code/linear_regressions.ipynb, lines 1445 to 1464 · score 0.61 · episodic memory, visual memory, global cognition, executive, AD, diagnostic · Report a mismatch
[4] § MATERIALS AND METHODS › Statistical analysis and predictor-based age estimation ↔ Code/loco_main.ipynb, lines 355 to 416 · score 0.57 · gradient boosting regressor, inner, R2, fold, predict, Model · Report a mismatch
[5] § MATERIALS AND METHODS › Brain clocks › Structural clock (MRI-T1) ↔ Code/regression_main.ipynb, lines 294 to 355 · score 0.56 · gradient boosting, outer, nested, inner, pipeline, fold · Report a mismatch
[6] § RESULTS › SAG building using speech composite features ↔ Code/loco_main.ipynb, lines 36 to 157 · score 0.56 · absolute error, squared error, MAE, MDE, predict · Report a mismatch
[7] § RESULTS › SAG building using speech composite features ↔ Code/regression_main.ipynb, lines 36 to 157 · score 0.56 · absolute error, squared error, MAE, MDE, predict · Report a mismatch
[8] § MATERIALS AND METHODS › Feature extraction (age predictors) ↔ Code/R2R/R2R-Fig3-prediction-power.ipynb, lines 199 to 263 · score 0.54 · processing speed, semantic memory, metric, predictive, age · Report a mismatch
[9] § RESULTS › Sensitivity models confirm diagnostic differences in SAGs ↔ Code/R2R/R2R-Fig-3-analysis.ipynb, lines 656 to 786 · score 0.54 · composite comorbidity, linguistic predictors, covariates, pitch, languages, education · Report a mismatch
[10] § RESULTS › SAGs outperform individual speech-feature domains in diagnostic discrimination ↔ Code/R2R/R2R-Fig3-prediction-power.ipynb, lines 199 to 263 · score 0.51 · processing speed, semantic memory, bootstrap, education, linguistic, sex · Report a mismatch

Paper

Loaded from Europe PMC by your browser; OSCR does not store it: 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

Jupyter notebook · 476 lines · 14 KB · no license · 3 matches

# %%
import pandas as pd
import numpy as np
import os
import pandas as pd
import numpy as np
from collections import Counter
import statsmodels.api as sm
from sklearn.model_selection import train_test_split
from sklearn.metrics import r2_score
from sklearn.model_selection import KFold
from sklearn.metrics import mean_squared_error, r2_score
import math
from sklearn.preprocessing import MinMaxScaler
import scipy
from sklearn import metrics
import numpy as np
from sklearn.metrics import mean_absolute_error
from scipy.stats import pearsonr
import seaborn as sns
import matplotlib.pyplot as plt
from tqdm.notebook import tqdm
from sklearn.linear_model import LinearRegression, Ridge
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, r2_score
from scipy.stats import linregress
from scipy import stats
from sklearn.utils import resample
# %%
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from scipy.stats import permutation_test
def perm_test(group1, group2, n_permutations=10000):
observed_diff = np.mean(group1) - np.mean(group2)
def statistic(x, y):
return np.mean(x) - np.mean(y)
result = permutation_test(
data=(group1, group2),
statistic=statistic,
permutation_type='independent',
vectorized=False, # importante: no usar vectorized
n_resamples=n_permutations,
alternative='two-sided',
random_state=42
)
return observed_diff, result.pvalue
def cliffs_delta(x, y):
n_x = len(x)
n_y = len(y)
all_pairs = [(xi, yi) for xi in x for yi in y]
n_greater = sum(1 for xi, yi in all_pairs if xi > yi)
n_less = sum(1 for xi, yi in all_pairs if xi < yi)
delta = (n_greater - n_less) / (n_x * n_y)
return delta
# Función para quitar outliers con IQR
def remove_outliers_iqr(df, col, thres=1.5):
Q1 = df[col].quantile(0.25)
Q3 = df[col].quantile(0.75)
IQR = Q3 - Q1
lower = Q1 - thres * IQR
upper = Q3 + thres * IQR
return df[(df[col] >= lower) & (df[col] <= upper)]
def mean_directional_accuracy(y_true, y_pred):
differences = np.array(y_pred) - np.array(y_true)
signs = np.sign(differences)
mde = np.mean(signs)
return mde
def mean_absolute_error(y_true, y_pred):
y_true = np.array(y_true)
y_pred = np.array(y_pred)
absolute_errors = np.abs(y_pred - y_true)
mae = np.mean(absolute_errors)
return mae
def coef_pval(coef_array_mean_, X_, y_, y_p):
n = X_.shape[0]
t = coef_tval(coef_array_mean_, X_, y_, y_p)
p = 2 * (1 - scipy.stats.t.cdf(abs(t), n - 1))
return p
def coef_tval(coef_array_mean_, X_, y_, y_p):
'''
coef_tval for OLS of statsmodels
'''
a = np.array(coef_array_mean_[0][0]/ coef_se(X_, y_, y_p)[0])
b = np.array(coef_array_mean_[1::].flatten() / coef_se(X_, y_, y_p)[1:])
return np.append(a, b)
def coef_se(X_, y_, y_p):
'''
coef_se for OLS of statsmodels
'''
n = X_.shape[0]
X1 = np.hstack((np.ones((n, 1)), np.matrix(X_)))
se_matrix = scipy.linalg.sqrtm(
metrics.mean_squared_error(y_, y_p) *
np.linalg.inv(X1.T * X1)
)
return np.diagonal(se_matrix)
def directional_accuracy(predicted_values, true_values):
predicted_values = np.array(predicted_values)
true_values = np.array(true_values)
difference_direction = np.sign(predicted_values - true_values)
correct_direction_count = np.sum(difference_direction == 1) + np.sum(difference_direction == -1)
directional_accuracy_score = correct_direction_count / len(predicted_values)
return directional_accuracy_score
from sklearn.ensemble import GradientBoostingRegressor
from statsmodels.api import GLM
from statsmodels.api import families
from sklearn.model_selection import GridSearchCV, KFold
from sklearn.ensemble import GradientBoostingRegressor
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import r2_score, mean_squared_error, mean_absolute_error
from scipy.stats import linregress
import numpy as np
import pandas as pd
import math
import warnings
from skopt import BayesSearchCV
from skopt.space import Real, Integer
# %%
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
def remove_outliers_iqr(df, col, iqr_k=1.5):
s = pd.to_numeric(df[col], errors="coerce")
q1 = s.quantile(0.25)
q3 = s.quantile(0.75)
iqr = q3 - q1
low = q1 - iqr_k * iqr
high = q3 + iqr_k * iqr
return df.loc[(s >= low) & (s <= high)].copy()
def cliffs_delta(x, y):
x = pd.to_numeric(pd.Series(x).dropna(), errors="coerce").dropna().values
y = pd.to_numeric(pd.Series(y).dropna(), errors="coerce").dropna().values
nx, ny = len(x), len(y)
if nx == 0 or ny == 0:
return np.nan
diffs = x[:, None] - y[None, :]
n_greater = np.sum(diffs > 0)
n_less = np.sum(diffs < 0)
return float((n_greater - n_less) / (nx * ny))
def permutation_pvalue(x, y, n_perm=10000, seed=42):
"""Two-sided permutation test for mean difference"""
rng = np.random.default_rng(seed)
x = np.array(x.dropna())
y = np.array(y.dropna())
if len(x) == 0 or len(y) == 0:
return np.nan
obs_diff = np.abs(np.mean(x) - np.mean(y))
combined = np.concatenate([x, y])
nx = len(x)
count = 0
for _ in range(n_perm):
rng.shuffle(combined)
x_perm, y_perm = combined[:nx], combined[nx:]
diff = np.abs(np.mean(x_perm) - np.mean(y_perm))
if diff >= obs_diff:
count += 1
return count / n_perm
def chunk_list(lst, n):
"""Split list into chunks of size n."""
for i in range(0, len(lst), n):
yield lst[i:i + n]
def plot_gap_by_diagnosis_clean_wrap(
df,
gap_col,
diagnosis_list,
thresholds=(0.1, 0.5, 1.0, 1.5),
per_line=3,
title_fontsize=8,
n_perm=10000
):
# Build original data by group
group_data_orig = {}
for diag in diagnosis_list:
sub = df[df["clinical_diagnosis"] == diag][[gap_col]].copy()
sub["Origen"] = diag
group_data_orig[diag] = sub
# Unique pairs without repetition
all_pairs = [(diagnosis_list[i], diagnosis_list[j])
for i in range(len(diagnosis_list))
for j in range(i+1, len(diagnosis_list))]
cn_pairs = [(a, b) for (a, b) in all_pairs if a == "CN"]
other_pairs = [p for p in all_pairs if p not in cn_pairs]
results_table = []
fig, axes = plt.subplots(2, 2, figsize=(7.0, 8.0), sharey=True)
axes = axes.flatten()
for ax, iqr_k in zip(axes, thresholds):
group_data_clean = {
diag: remove_outliers_iqr(group_data_orig[diag], gap_col, iqr_k)
for diag in diagnosis_list
}
cn_lines = []
for a, b in cn_pairs:
x = group_data_clean[a][gap_col]
y = group_data_clean[b][gap_col]
dval = cliffs_delta(x, y)
pval = permutation_pvalue(x, y, n_perm=n_perm)
cn_lines.append(f"{a}-{b}: {dval:.3f}")
results_table.append([iqr_k, a, b, dval, pval])
other_lines = []
for a, b in other_pairs:
x = group_data_clean[a][gap_col]
y = group_data_clean[b][gap_col]
dval = cliffs_delta(x, y)
pval = permutation_pvalue(x, y, n_perm=n_perm)
other_lines.append(f"{a}-{b}: {dval:.3f}")
results_table.append([iqr_k, a, b, dval, pval])
cn_wrapped = [" | ".join(chunk) for chunk in chunk_list(cn_lines, per_line)]
other_wrapped = [" | ".join(chunk) for chunk in chunk_list(other_lines, per_line)]
title_parts = [f"IQR={iqr_k}"] + cn_wrapped + other_wrapped
ax.set_title("\n".join(title_parts), fontsize=title_fontsize)
combined = pd.concat([group_data_clean[d] for d in diagnosis_list], ignore_index=True)
sns.boxplot(data=combined, x="Origen", y=gap_col, order=diagnosis_list, ax=ax)
ax.axhline(0, color='gray', linestyle='--', linewidth=1)
ax.set_xlabel("")
ax.set_ylabel(gap_col)
plt.tight_layout()
results_df = pd.DataFrame(results_table, columns=["IQR_k", "Group1", "Group2", "CliffsDelta", "p_value"])
return fig, axes, results_df
# %%
import numpy as np
import pandas as pd
from sklearn.model_selection import KFold, GridSearchCV
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import MinMaxScaler
from sklearn.ensemble import HistGradientBoostingRegressor
from sklearn.metrics import r2_score
from sklearn.inspection import permutation_importance
from scipy.stats import linregress
import shap
def run_nested_hgb(
df_data,
selected_hgb,
n_splits_outer = 10,
n_splits_inner = 5,
random_state = 42,
scaler_range=(0.05, 0.95),
param_grid=None,
n_perm_repeats: int = 30,
shap_check_additivity: bool = False,
verbose: int = 2
):
if param_grid is None:
param_grid = {
"model__max_iter": [300, 400, 500, 600],
"model__max_depth": [3, 5, 7],
"model__learning_rate": [0.01, 0.05, 0.1],
}
if isinstance(selected_hgb, str):
selected_hgb = [selected_hgb]
selected_hgb = list(selected_hgb)
y = df_data["demo_age"]
X_selected = df_data[selected_hgb]
kf_outer = KFold(n_splits=n_splits_outer, shuffle=True, random_state=random_state)
r2_scores = []
y_true_all, y_pred_all = [], []
results_labels_df = pd.DataFrame(columns=[
'y_labels', 'y_pred', 'y_pred_M1', 'y_pred_M2', 'y_pred_M3',
'GAP', 'GAP_corrected_M1', 'GAP_corrected_M2', 'GAP_corrected_M3', 'ID'
])
shap_values_sum = pd.Series(0.0, index=X_selected.columns)
perm_values_sum = pd.Series(0.0, index=X_selected.columns)
best_params_by_fold = []
for fold, (train_idx, test_idx) in enumerate(kf_outer.split(X_selected), start=1):
print(f" Fold {fold} (Nested CV)")
X_train, X_test = X_selected.iloc[train_idx], X_selected.iloc[test_idx]
y_train, y_test = y.iloc[train_idx], y.iloc[test_idx]
pipeline = Pipeline([
("scaler", MinMaxScaler(feature_range=scaler_range)),
("model", HistGradientBoostingRegressor(random_state=random_state))
])
grid_search = GridSearchCV(
estimator=pipeline,
param_grid=param_grid,
scoring="r2",
cv=n_splits_inner,
n_jobs=-1,
verbose=verbose
)
grid_search.fit(X_train, y_train)
best_model = grid_search.best_estimator_
best_params_by_fold.append(grid_search.best_params_)
y_pred = best_model.predict(X_test)
r2 = r2_score(y_test, y_pred)
r2_scores.append(r2)
print(f" R² fold {fold}: {r2:.4f} (best params: {grid_search.best_params_})")
y_true_all.extend(y_test.tolist())
y_pred_all.extend(y_pred.tolist())
gap_test = y_pred - y_test
gap_train_all = best_model.predict(X_train) - y_train
train_ids = X_train.index
diag_train = df_data.loc[train_ids, 'clinical_diagnosis']
cn_mask = (diag_train == 'CN').values
if cn_mask.sum() >= 2:
slope, intercept, _, _, _ = linregress(y_train.values[cn_mask], gap_train_all.values[cn_mask])
else:
slope, intercept = 0.0, 0.0
corrected_gap_M1 = gap_test - (slope * y_test + intercept)
y_pred_M1 = y_test + corrected_gap_M1
y_train_pred = best_model.predict(X_train)
if cn_mask.sum() >= 2:
slope_pred, intercept_pred, _, _, _ = linregress(y_train.values[cn_mask], y_train_pred[cn_mask])
if slope_pred == 0:
slope_pred = 1.0
else:
slope_pred, intercept_pred = 1.0, 0.0
y_pred_M2 = (y_pred - intercept_pred) / slope_pred
corrected_gap_M2 = y_pred_M2 - y_test
y_pred_M3 = (y_pred - intercept_pred) / slope_pred
corrected_gap_M3 = y_pred_M3 - y_test
explainer = shap.Explainer(best_model.named_steps['model'], X_train)
shap_values = explainer(X_test, check_additivity=shap_check_additivity)
shap_values_mean = np.abs(shap_values.values).mean(axis=0)
shap_values_sum += pd.Series(shap_values_mean, index=X_selected.columns)
perm_importance = permutation_importance(
best_model, X_test, y_test,
n_repeats=n_perm_repeats,
random_state=random_state,
n_jobs=-1
)
perm_values_sum += pd.Series(perm_importance.importances_mean, index=X_selected.columns)
result = np.column_stack((
y_test, y_pred, y_pred_M1, y_pred_M2, y_pred_M3,
gap_test, corrected_gap_M1, corrected_gap_M2, corrected_gap_M3
))
temp_df = pd.DataFrame(result, columns=[
'y_labels', 'y_pred', 'y_pred_M1', 'y_pred_M2', 'y_pred_M3',
'GAP', 'GAP_corrected_M1', 'GAP_corrected_M2', 'GAP_corrected_M3'
])
temp_df['ID'] = X_test.index
results_labels_df = pd.concat([results_labels_df, temp_df], ignore_index=True)
r2_mean = float(np.mean(r2_scores))
r2_std = float(np.std(r2_scores))
r2_global = float(r2_score(y_true_all, y_pred_all))
print("\n R² promedio:", r2_mean, r2_std)
print(" R² global:", r2_global)
shap_importance_avg = (shap_values_sum / kf_outer.get_n_splits()).sort_values(ascending=False)
perm_importance_avg = (perm_values_sum / kf_outer.get_n_splits()).sort_values(ascending=False)
results_labels_df = results_labels_df.set_index('ID').sort_index()
df_final = pd.concat(
[df_data[['demo_age', 'clinical_diagnosis'] + selected_hgb], results_labels_df],
axis=1
)
out = {
"selected_hgb": selected_hgb,
"df_final": df_final,
"results_labels_df": results_labels_df,
"r2_scores": r2_scores,
"r2_mean": r2_mean,
"r2_std": r2_std,
"r2_global": r2_global,
"y_true_all": y_true_all,
"y_pred_all": y_pred_all,
"best_params_by_fold": best_params_by_fold,
"shap_importance_avg": shap_importance_avg,
"perm_importance_avg": perm_importance_avg,
}
return out
# %% [markdown]
# ## Load Data
# %%
df_data = pd.read_excel('../Data/augmented_data.xlsx')
# %%
selected_hgb = ['concreteness',
'granularity_extraction', 'psycholinguistic_objective',
'pitch_analysis', 'sentiment_analysis', 'talking_intervals',
'verbosity']
# %% [markdown]
# ## Regression
# %%
results = run_nested_hgb(df_data, selected_hgb, n_splits_outer = 10, n_splits_inner = 5)
# %%

regression_main.ipynb at commit d75d5c3, no license · at the source

Overview

Authors: Hernan Hernandez¹, Lizeth Katherine Pedraza², Hernando Santamaria-Garcia^3,4,5,6,7, Sebastian Moguilner^1,8,9, Agustina Legaz^1,8,10, Pavel Prado¹¹, Jhosmary Cuadros^1,12,13, Lucía Amoruso^8,14,15, Liset Gonzalez^1,16, Damián Dellavale^17,18, Juan Pablo Espinoza–Puelles^1,19,20, Javier Palma Espinosa^1,21, Cecilia Jarne^1,22,23,24, Fabio Mattiussi^1,25, Matías Caccia⁸, Alejandro Sosa Welford^8,22, Nicolás Pelella⁸, Jeremías Inchauspe^8,22, Franco J Ferrante^8,26,27, Gonzalo Pérez^8,26,27

and 32 other authors

Marcelo Adrián Maito^1,8, Guido Rocatti⁸, Maria Eugenia Godoy^1,8,18, Joaquin Migeot^1,18, Paulina Orellana¹, Ariel Caviedes¹, Martin Bruno^26,28, Leonel Takada²⁹, Andrea Slachevsky^30,31,32,33, Maria I Behrens^33,34,35,36, Bárbara Bruna³⁴, David Aguillon³⁷, Lina Zapata^10,38,39, Jose Alberto Avila-Funes⁴⁰, Nilton Custodio⁴¹, Bruce Miller^10,42, Maria Luisa Gorno-Tempini^10,43, Stefanie Pina Escudero^10,44, Pablo Reyes^45,46, Kun Hu^47,48, Maira Okada de Oliveira^10,49, Carlos Coronel-Oliveros^1,43,50, Josephine Cruzat¹, Juan Felipe Cardona⁵¹, Michael Corley^52,53, Irene B Meier⁵⁴, Vaibhav A Narayan^54,55, Enzo Tagliazucchi^1,56, Sandra Baez^18,57, Claudia Duran-Aniotz¹, Adolfo M García^8,10,58, Agustin Ibanez^1,8,43,59,60

60 affiliations

Institutions: Adolfo Ibáñez University (Chile); Universidad de La Sabana (Colombia); Universidad de Los Andes (Colombia); Pontificia Universidad Javeriana (Colombia); Universidad Nacional de Colombia (Colombia); Fundación Santa Fe de Bogotá (Colombia); Hospital Universitario San Ignacio (Colombia); Harvard University (United States); University of San Andrés (Argentina); University of California, San Francisco (United States); Global Brain Health Institute (United States); Universidad San Sebastián (Chile); Universidad Nacional Experimental del Táchira (Venezuela); Universidad Técnica Federico Santa María (Chile); Ikerbasque (Spain); Basque Center on Cognition, Brain and Language (Spain); University of Concepción (Chile); Consejo Nacional de Investigaciones Científicas y Técnicas (Argentina); Trinity College Dublin (Ireland); Bariloche Atomic Centre (Argentina); Pontificia Universidad Católica de Valparaíso (Chile); National University of Quilmes (Argentina); Aarhus University Hospital (Denmark); Universitat Pompeu Fabra (Spain); Universidad de Buenos Aires (Argentina); Universidad Católica de Cuyo (Argentina); Universidade de São Paulo (Brazil); Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo (Brazil); Universidad del Desarrollo (Chile); Clínica Alemana (Chile); Hospital del Salvador (Chile); Instituto de Neurociencia Biomédica (Chile); University of Chile (Chile); Hospital Clínico de la Universidad de Chile (Chile); Universidad de Antioquia (Colombia); Fundación Valle del Lili (Colombia); Universidad Icesi (Colombia); Instituto Nacional de Ciencias Médicas y Nutrición Salvador Zubirán (Mexico); Institute of Peruvian Studies (Peru); University Memory and Aging Center (United States); Massachusetts General Hospital (United States); Universidad del Valle (Colombia); University of California San Diego (United States); Institute on Aging (United States); Alzheimer’s Disease Neuroimaging Initiative (United States); Universidad de Santiago de Chile (Chile); Istanbul Medipol University (Türkiye); Pasqual Maragall Foundation (Spain); Barcelonaβeta Brain Research Center (Spain)

Journal: Science advances, volume 12, issue 40, article eaef9864

Dates: received 2 February 2026; accepted 25 August 2026; published online 30 September 2026; in print October 2026

Type: Research article · Language: English

License: CC BY

Identifiers: DOI 10.1126/sciadv.aef9864 · PMID 42814823 · PMCID PMC13626050 · OpenAlex W7214974072

Open access: gold, a free copy (OpenAlex)

Status: code verified (how the code is found and verified)

Categories: human (organism), Alzheimer's / dementia (population), cognitive (subfield)

Methods: Spectral & time-frequency, Connectivity, Statistics, Machine learning, Preprocessing, fMRI & imaging, Physiology & signal measures

MeSH: Aging*, Dementia*, Frontotemporal Dementia*, Speech*, Aged, Alzheimer Disease, Cognitive Dysfunction, Female, Humans, Male, Phenotype (* major topic)

Topic: Dementia and Cognitive Impairment Research (Psychiatry and Mental health, Medicine), according to OpenAlex

Funding: Alzheimer's Association (SG-20- 725707, ANID/FOVI240065, ALZ-RWD-26-1466627, AARGD-24-1310017); Wellcome Trust (227012/Z/23/Z, 335293/Z/25/Z); Global Brain Health Institute (GBHI ALZ UK-23-971135); Rainwater Charitable Foundation; Alzheimer's Society; Ministerio de Ciencia y Tecnología (RYC2022-035514-I); Agencia Nacional de Investigación y Desarrollo (13240170, ANID/PIA/ANILLOS ACT210096, 21210855, 15150012, 3240042, FONDEF ID20I10152, ANID/FONDAP/15150012, NAM22I0007, 21250152, 1210622, 1250091); National Institutes of Health (1r01ag082056-01a1, R01AG21051, R01AG083799, 2P01AG019724, R01AG057234, R01AG075775, GBHI ALZ UK-23-971135, 1d43tw012455-01a1); University of California, San Francisco (P0544014); Fondo para la Investigación Científica y Tecnológica (01‐PICTE‐2022‐05‐00103); Fogarty International Center (ReDLat2); EQuity Social Venture (NIH D43 F, R56AG069118-01, SG-21-715176-LATAM FIN- GERS, 24AARG-D-1246942, DOCTORADO BECAS CHILE/2022, SG-21-715176-LATAM FIN, GBHI ALZ UK-25-1289623, R01AG082056, 24AARG-D-1246942); Wellcome Leap (CARE-2025-0883490149)

Citations: not cited yet (Europe PMC); 68 references in the paper

Abstract

Biological aging clocks offer estimations of aging and dementia, yet scalability is limited. We introduce a large-scale, cross-national speech clock derived from 2928 individuals across five Latin American countries, spanning healthy controls (HCs), mild cognitive impairment (MCI), Alzheimer’s disease (AD), and non-language/language-dominant frontotemporal dementia (nldFTD/ldFTD). Multimodal acoustic and linguistic features were trained with supervised models to estimate chronological age, generating speech age gaps (SAGs) as cross-sectional markers of deviations from chronological age, with positive values interpreted as relatively older-appearing speech profiles. SAGs differentiated diagnostic groups (HCs < patient groups, with AD < nldFTD < ldFTD). This pattern was associated with clinical/cognitive domains. SAGs correlated with phosphorylated tau (p-Tau217) in AD and social exposome in HCs and AD. Brain clocks (structural/functional/combined) were associated with SAG in AD, nldFTD, and ldFTD. Epigenetic age correlated with SAGs in HCs and AD across Hannum, Retroclock, and OMICmAge, whereas ldFTD associations were limited to Retroclock and OMICmAge. These results indicate that SAGs capture cross-sectional multilevel aging-related variation and may offer a scalable, culturally adaptable, low-cost biomarker candidate for research in underrepresented global settings.

Reproduced under the paper's license (CC BY), from the paper cited above.

Repositories

Its files are read in the Code ↔ Paper reader above, with 10 matches between paragraphs and lines of code.

Zenodo 20186949

License: CC-BY-4.0

State: the link answers, verified on 3 October 2026

Evidence: files inventoried

Languages: Jupyter (13)

Size: 35 files, 13 scripts

Software Heritage: not checked

Found in: “Data, code, and materials availability:”

Holds: 13 notebooks

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

Tools: NumPy (13 files), pandas (13 files), SciPy (13 files), statsmodels (13 files), Matplotlib (11 files), seaborn (9 files), scikit-learn (7 files), SHAP (4 files)

Availability: 1 check, the latest on 3 October 2026: the link answers (HTTP 200)

3 October 2026: the link answers (HTTP 200)

13 files

R2R-Fig-3-analysis-checkpoint.ipynb, Jupyter, 1,089 lines
R2R-Fig-3-analysis.ipynb, Jupyter, 1,089 lines
R2R-Fig3-prediction-power-checkpoint.ipynb, Jupyter, 271 lines
R2R-Fig3-prediction-power.ipynb, Jupyter, 271 lines
R2R-Reviewer-1-powerOfLinguisticFeat-checkpoint.ipynb, Jupyter, 1,581 lines
Stats-checkpoint.ipynb, Jupyter, 75 lines
Stats.ipynb, Jupyter, 75 lines
linear_regressions-checkpoint.ipynb, Jupyter, 1,569 lines
linear_regressions.ipynb, Jupyter, 1,569 lines
loco_main-checkpoint.ipynb, Jupyter, 493 lines
loco_main.ipynb, Jupyter, 493 lines
regression_main-checkpoint.ipynb, Jupyter, 476 lines
regression_main.ipynb, Jupyter, 476 lines

At the source: doi.org/10.5281/zenodo.20186949

euroladbrainlat/Speech-age-gap

License: none: the authors keep all their rights

State: the link answers, verified on 3 October 2026

Evidence: files inventoried

Commit: d75d5c3f1a713c9c5a501e900be7af33fb38b0bf, 18 May 2026

Languages: Jupyter (6)

Size: 30 files, 6 scripts

Software Heritage: not archived

Found in: “Data, code, and materials availability:”

Holds: README, 6 notebooks

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

Tools: NumPy (6 files), pandas (6 files), SciPy (6 files), statsmodels (6 files), Matplotlib (5 files), seaborn (4 files), scikit-learn (3 files), SHAP (2 files)

Availability: 1 check, the latest on 3 October 2026: the link answers

3 October 2026: the link answers

7 files

Code/R2R/R2R-Fig-3-analysis.ipynb, Jupyter, 1,089 lines, 1 match
Code/R2R/R2R-Fig3-prediction-power.ipynb, Jupyter, 271 lines, 2 matches
Code/Stats.ipynb, Jupyter, 75 lines
Code/linear_regressions.ipynb, Jupyter, 1,569 lines, 2 matches
Code/loco_main.ipynb, Jupyter, 493 lines, 2 matches
Code/regression_main.ipynb, Jupyter, 476 lines, 3 matches
README.md, Text, 1 line

At the source: github.com/euroladbrainlat/Speech-age-gap

The paper's code and data availability statement is in the Data section.

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:

2 repositories of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
19 scripts, each with its path and the digest of its content;
10 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

No dataset and no data link were found in the paper.

Data, code, and materials availability

All data and code needed to evaluate and reproduce the results in the paper are present in the paper and/or the Supplementary Materials (https://doi.org/10.5281/zenodo.20186949 and https://github.com/euroladbrainlat/Speech-age-gap). This study did not generate new materials. Due to regulatory and ethical considerations, the underlying recorded datasets are available under restricted access, as the acoustic and linguistic features of the recorded and transcribed materials contain identifiable information about the participants. Access can be obtained after IRB approval of the formal data-sharing agreement in a process that can last up to 12 weeks. For questions related to the data request and usage, contact A.I. at .

Reproduced under the paper's license (CC BY), from the paper cited above.

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, 3 October 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 52 authors, 11 MeSH terms, 13 funders, 65 references.

Cite

This paper

Hernandez, H., Pedraza, L. K., Santamaria-Garcia, H., Moguilner, S., Legaz, A., Prado, P., Cuadros, J., Amoruso, L., Gonzalez, L., Dellavale, D., Espinoza–Puelles, J. P., Espinosa, J. P., Jarne, C., Mattiussi, F., Caccia, M., Welford, A. S., Pelella, N., Inchauspe, J., Ferrante, F. J., . . . Ibanez, A. (2026). Speech clocks decode dementia phenotypes, social exposome, and biological aging. Science advances, 12(40), eaef9864. https://doi.org/10.1126/sciadv.aef9864

BibTeX

@article{hernandez2026speech, author = {Hernandez, Hernan and Pedraza, Lizeth Katherine and Santamaria-Garcia, Hernando and Moguilner, Sebastian and Legaz, Agustina and Prado, Pavel and Cuadros, Jhosmary and Amoruso, Lucía and Gonzalez, Liset and Dellavale, Damián and Espinoza–Puelles, Juan Pablo and Espinosa, Javier Palma and Jarne, Cecilia and Mattiussi, Fabio and Caccia, Matías and Welford, Alejandro Sosa and Pelella, Nicolás and Inchauspe, Jeremías and Ferrante, Franco J and Pérez, Gonzalo and Maito, Marcelo Adrián and Rocatti, Guido and Godoy, Maria Eugenia and Migeot, Joaquin and Orellana, Paulina and Caviedes, Ariel and Bruno, Martin and Takada, Leonel and Slachevsky, Andrea and Behrens, Maria I and Bruna, Bárbara and Aguillon, David and Zapata, Lina and Avila-Funes, Jose Alberto and Custodio, Nilton and Miller, Bruce and Gorno-Tempini, Maria Luisa and Escudero, Stefanie Pina and Reyes, Pablo and Hu, Kun and de Oliveira, Maira Okada and Coronel-Oliveros, Carlos and Cruzat, Josephine and Cardona, Juan Felipe and Corley, Michael and Meier, Irene B and Narayan, Vaibhav A and Tagliazucchi, Enzo and Baez, Sandra and Duran-Aniotz, Claudia and García, Adolfo M and Ibanez, Agustin}, title = {{Speech clocks decode dementia phenotypes, social exposome, and biological aging}}, journal = {Science advances}, year = {2026}, month = sep, volume = {12}, number = {40}, pages = {eaef9864}, publisher = {American Association for the Advancement of Science}, issn = {2375-2548}, doi = {10.1126/sciadv.aef9864}, url = {https://doi.org/10.1126/sciadv.aef9864}, pmid = {42814823}, pmcid = {PMC13626050} }

RIS

TY - JOUR AU - Hernandez, Hernan AU - Pedraza, Lizeth Katherine AU - Santamaria-Garcia, Hernando AU - Moguilner, Sebastian AU - Legaz, Agustina AU - Prado, Pavel AU - Cuadros, Jhosmary AU - Amoruso, Lucía AU - Gonzalez, Liset AU - Dellavale, Damián AU - Espinoza–Puelles, Juan Pablo AU - Espinosa, Javier Palma AU - Jarne, Cecilia AU - Mattiussi, Fabio AU - Caccia, Matías AU - Welford, Alejandro Sosa AU - Pelella, Nicolás AU - Inchauspe, Jeremías AU - Ferrante, Franco J AU - Pérez, Gonzalo AU - Maito, Marcelo Adrián AU - Rocatti, Guido AU - Godoy, Maria Eugenia AU - Migeot, Joaquin AU - Orellana, Paulina AU - Caviedes, Ariel AU - Bruno, Martin AU - Takada, Leonel AU - Slachevsky, Andrea AU - Behrens, Maria I AU - Bruna, Bárbara AU - Aguillon, David AU - Zapata, Lina AU - Avila-Funes, Jose Alberto AU - Custodio, Nilton AU - Miller, Bruce AU - Gorno-Tempini, Maria Luisa AU - Escudero, Stefanie Pina AU - Reyes, Pablo AU - Hu, Kun AU - de Oliveira, Maira Okada AU - Coronel-Oliveros, Carlos AU - Cruzat, Josephine AU - Cardona, Juan Felipe AU - Corley, Michael AU - Meier, Irene B AU - Narayan, Vaibhav A AU - Tagliazucchi, Enzo AU - Baez, Sandra AU - Duran-Aniotz, Claudia AU - García, Adolfo M AU - Ibanez, Agustin TI - Speech clocks decode dementia phenotypes, social exposome, and biological aging T2 - Science advances J2 - Sci Adv PY - 2026 DA - 2026/09/30 VL - 12 IS - 40 SP - eaef9864 SN - 2375-2548 PB - American Association for the Advancement of Science DO - 10.1126/sciadv.aef9864 UR - https://doi.org/10.1126/sciadv.aef9864 LA - en ER -

CSL-JSON

{ "id": "10.1126/sciadv.aef9864", "type": "article-journal", "title": "Speech clocks decode dementia phenotypes, social exposome, and biological aging", "container-title": "Science advances", "author": [ { "family": "Hernandez", "given": "Hernan" }, { "family": "Pedraza", "given": "Lizeth Katherine" }, { "family": "Santamaria-Garcia", "given": "Hernando" }, { "family": "Moguilner", "given": "Sebastian" }, { "family": "Legaz", "given": "Agustina" }, { "family": "Prado", "given": "Pavel" }, { "family": "Cuadros", "given": "Jhosmary" }, { "family": "Amoruso", "given": "Lucía" }, { "family": "Gonzalez", "given": "Liset" }, { "family": "Dellavale", "given": "Damián" }, { "family": "Espinoza–Puelles", "given": "Juan Pablo" }, { "family": "Espinosa", "given": "Javier Palma" }, { "family": "Jarne", "given": "Cecilia" }, { "family": "Mattiussi", "given": "Fabio" }, { "family": "Caccia", "given": "Matías" }, { "family": "Welford", "given": "Alejandro Sosa" }, { "family": "Pelella", "given": "Nicolás" }, { "family": "Inchauspe", "given": "Jeremías" }, { "family": "Ferrante", "given": "Franco J" }, { "family": "Pérez", "given": "Gonzalo" }, { "family": "Maito", "given": "Marcelo Adrián" }, { "family": "Rocatti", "given": "Guido" }, { "family": "Godoy", "given": "Maria Eugenia" }, { "family": "Migeot", "given": "Joaquin" }, { "family": "Orellana", "given": "Paulina" }, { "family": "Caviedes", "given": "Ariel" }, { "family": "Bruno", "given": "Martin" }, { "family": "Takada", "given": "Leonel" }, { "family": "Slachevsky", "given": "Andrea" }, { "family": "Behrens", "given": "Maria I" }, { "family": "Bruna", "given": "Bárbara" }, { "family": "Aguillon", "given": "David" }, { "family": "Zapata", "given": "Lina" }, { "family": "Avila-Funes", "given": "Jose Alberto" }, { "family": "Custodio", "given": "Nilton" }, { "family": "Miller", "given": "Bruce" }, { "family": "Gorno-Tempini", "given": "Maria Luisa" }, { "family": "Escudero", "given": "Stefanie Pina" }, { "family": "Reyes", "given": "Pablo" }, { "family": "Hu", "given": "Kun" }, { "family": "de Oliveira", "given": "Maira Okada" }, { "family": "Coronel-Oliveros", "given": "Carlos" }, { "family": "Cruzat", "given": "Josephine" }, { "family": "Cardona", "given": "Juan Felipe" }, { "family": "Corley", "given": "Michael" }, { "family": "Meier", "given": "Irene B" }, { "family": "Narayan", "given": "Vaibhav A" }, { "family": "Tagliazucchi", "given": "Enzo" }, { "family": "Baez", "given": "Sandra" }, { "family": "Duran-Aniotz", "given": "Claudia" }, { "family": "García", "given": "Adolfo M" }, { "family": "Ibanez", "given": "Agustin" } ], "container-title-short": "Sci Adv", "volume": "12", "issue": "40", "page": "eaef9864", "DOI": "10.1126/sciadv.aef9864", "PMID": "42814823", "PMCID": "PMC13626050", "ISSN": "2375-2548", "publisher": "American Association for the Advancement of Science", "URL": "https://doi.org/10.1126/sciadv.aef9864", "language": "en", "issued": { "date-parts": [ [ 2026, 9, 30 ] ] } }

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

Similar papers

The papers with a page that share the most with this one: the tools found in their code, their categories, datasets, cited references and authors, the rarest counting most.

[1] doi:10.1038/s41591-024-03209-x [code]: Brain clocks capture diversity and disparities in aging and dementia across geographically diverse populations.
Journal: Nature medicine
In common: Alzheimer's / dementia, 10 references, 14 authors
[2] doi:10.1038/s41467-025-63277-6 [code]: Social exposome and brain health outcomes of dementia across Latin America.
Journal: Nature communications
In common: Alzheimer's / dementia, 13 references, 9 authors
[3] doi:10.1002/alz.14085 [code]: Educational disparities in brain health and dementia across Latin America and the United States.
Journal: Alzheimer's & dementia : the journal of the Alzheimer's Association
In common: Alzheimer's / dementia, 10 references, 9 authors
[4] doi:10.1016/j.lana.2022.100387: Classification of Alzheimer's disease and frontotemporal dementia using routine clinical and cognitive measures across multicentric underrepresented samples: A cross sectional observational study.
Journal: Lancet regional health. Americas
In common: Alzheimer's / dementia, 3 references, 10 authors
[5] doi:10.1093/cvr/cvaf185: Cardiovascular risk factors and the allostatic interoceptive network in dementia.
Journal: Cardiovascular research
In common: Alzheimer's / dementia, 7 references, 8 authors
[6] doi:10.1038/s41597-023-02806-8 [code]: The BrainLat project, a multimodal neuroimaging dataset of neurodegeneration from underrepresented backgrounds.
Journal: Scientific data
In common: SciPy, NumPy, Alzheimer's / dementia, 7 references, 7 authors
[7] doi:10.1016/j.eclinm.2025.103187: Qualitative and quantitative educational disparities and brain signatures in healthy aging and dementia across global settings.
Journal: EClinicalMedicine
In common: Alzheimer's / dementia, 4 references, 7 authors
[8] doi:10.1038/s41467-025-64173-9 [code]: Creative experiences and brain clocks.
Journal: Nature communications
In common: statsmodels, seaborn, scikit-learn, 4 other tools, cognitive, 7 references, 4 authors
[9] doi:10.1186/s13195-025-01832-5: White matter hyperintensities and their impact in brain structure and function in alzheimer's disease and behavioral variant frontotemporal dementia across Latin America and the United States: a cross-sectional study.
Journal: Alzheimer's research & therapy
In common: Alzheimer's / dementia, 7 references, 5 authors
[10] doi:10.1002/alz.71232: Neural embedding of frailty in cognitively unimpaired aging and dementia across Latin America.
Journal: Alzheimer's & dementia : the journal of the Alzheimer's Association
In common: Alzheimer's / dementia, 11 references, 4 authors

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.

Add the badge to the code's README

If you are an author or a maintainer of this paper's code, you can add this badge. It says what OSCR knows of the paper (its code, its validated tracing map) and links the code to this page. Paste one of these snippets under the README's title. Keep the two marks around it, so that the badge can be replaced in place and is never added twice. OSCR changes nothing in your repository.

Markdown

<!-- oscr-badge:start -->
[![OSCR](https://openscicode.org/badge/doi/10.1126/sciadv.aef9864.svg)](https://openscicode.org/badge/doi/10.1126/sciadv.aef9864)
<!-- oscr-badge:end -->

reStructuredText

.. oscr-badge:start

.. image:: https://openscicode.org/badge/doi/10.1126/sciadv.aef9864.svg
   :target: https://openscicode.org/badge/doi/10.1126/sciadv.aef9864
   :alt: OSCR

.. oscr-badge:end

HTML

<!-- oscr-badge:start -->
<a href="https://openscicode.org/badge/doi/10.1126/sciadv.aef9864"><img src="https://openscicode.org/badge/doi/10.1126/sciadv.aef9864.svg" alt="OSCR"></a>
<!-- oscr-badge:end -->

On GitHub, open the README, choose Edit (the pencil), paste the snippet under the title, then "Commit changes".

euroladbrainlat/Speech-age-gap: README.md on GitHub

If you choose to, you can have the badge kept up to date for you, with oscr badge --write from the command line or with the GitHub Action in your repository's workflows. Every other form and option (a card, AsciiDoc, LaTeX, shields.io) is on the badge 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: code errors and mismatches between the code and the paper about this paper and its code, from readers and its authors, beside the paper and its tracing map.

Report a mismatch between the code and the paper · Report a code error (signed in)

Reproductions: reports from readers who ran the authors' code and did not get the paper's result, what they ran, with which environment, commit and data.

Report a reproduction failure (signed in)

Versions of its code: the releases of its code tied to a version of this paper (the preprint, the accepted manuscript, the version of record…), each with the tracing map versioned with it.

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