OSCR

Neuronal tuning aligns dynamically with object and texture manifolds across the visual hierarchy.

Code ↔ Paper

9 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 9 matches
  1. [1] § Methods › Tuning landscapes › Statistical analysis ↔ figure_reproduction/Figure6_reproduction.py, lines 183–205 · score 0.77 · Gaussian curve fitting, Gaussian process regression, linear regression, Tuning curves
  2. [2] § Results › Charting tuning landscapes in the BigGAN latent space ↔ figure_reproduction/Figure6_reproduction.py, lines 1–26 · score 0.77 · tuning curve peak, class space, latent space, Neuronal responses, eigenvector, bell
  3. [3] § Results › Charting tuning landscapes in the BigGAN latent space ↔ figure_reproduction/Figure6_reproduction.py, lines 1–26 · score 0.75 · tuning landscape, latent space, neuronal responses, neuronal tuning, bell, ramp
  4. [4] § Results › Alignment as a facility of hill climbing ↔ figure_reproduction/Figure4_reproduction.py, lines 1–19 · score 0.69 · ventral hierarchy, achieved activation, evolution success rate, evolution trajectory, Alignment, DeePSim
  5. [5] § Results › Neuron-guided image synthesis in two generative spaces ↔ figure_reproduction/Figure3_reproduction.py, lines 1–17 · score 0.66 · shared local motifs, neuron responded, Optimized images, global, texture, PSTHs
  6. [6] § Methods › Analysis of dynamic neuronal responses › Temporal attribution of activation changes ↔ source_data_export/Figure5_source_data_export.ipynb, lines 101–143 · score 0.64 · 0–200 ms, post stimulus, window, activation
  7. [7] § Methods › Analysis of dynamic neuronal responses › Time-window-specific evolutionary trajectory ↔ source_data_export/Figure5_source_data_export.ipynb, lines 101–143 · score 0.58 · 0–200 ms, DeePSim, windows, bin, BigGAN, trajectories
  8. [8] § Results › Neuron-guided image synthesis in two generative spaces ↔ figure_reproduction/Figure2_reproduction.py, lines 1–17 · score 0.56 · response latencies, PIT multi unit, firing rate, DeePSim, BigGAN, texture
  9. [9] § Results ↔ figure_reproduction/Figure4_reproduction.py, lines 1–19 · score 0.52 · ventral hierarchy, visual hierarchy, DeePSim, alignment, BigGAN, dynamics

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 authors' code

Python · 301 lines · 16 KB · MIT · 3 matches

  1. """
  2. Reproduction script for Figure 6 of:
  3. "Neuronal tuning aligns dynamically with object and texture manifolds across the visual hierarchy"
  4. Nature Neuroscience, 2026
  6. Figure 6: Geometry of tuning landscapes in BigGAN latent space (Hessian analysis).
  7. - Panel 6C: Example bell-shaped tuning curve along a sampled class-space eigenvector axis
  8. - Panel 6D: Heatmap of mean neuronal responses along multiple eigenvectors (class and noise subspace)
  9. - Panel 6E: Distribution of tuning curve peak locations and shape types (bell-shaped vs. ramp)
  10. as a function of BigGAN evolution success
  11. """
  12. import math
  13. import os
  14. import pandas as pd
  15. from os.path import join
  16. import numpy as np
  17. import seaborn as sns
  18. import matplotlib.pyplot as plt
  19. from matplotlib.colors import ListedColormap
  20. from scipy.optimize import curve_fit
  21. from sklearn.linear_model import LinearRegression
  22. from sklearn.gaussian_process import GaussianProcessRegressor
  23. from sklearn.gaussian_process.kernels import RBF, WhiteKernel, ConstantKernel
  24. import statsmodels.api as sm
  25. from statsmodels.formula.api import ols
  26. from statsmodels.stats.proportion import proportion_confint
  28. # --- Parula colormap (MATLAB default) ---
  29. _parula_data = [
  30. [0.2081, 0.1663, 0.5292], [0.2116238095, 0.1897809524, 0.5776761905],
  31. [0.212252381, 0.2137714286, 0.6269714286], [0.2081, 0.2386, 0.6770857143],
  32. [0.1959047619, 0.2644571429, 0.7279], [0.1707285714, 0.2919380952, 0.779247619],
  33. [0.1252714286, 0.3242428571, 0.8302714286], [0.0591333333, 0.3598333333, 0.8683333333],
  34. [0.0116952381, 0.3875095238, 0.8819571429], [0.0059571429, 0.4086142857, 0.8828428571],
  35. [0.0165142857, 0.4266, 0.8786333333], [0.032852381, 0.4430428571, 0.8719571429],
  36. [0.0498142857, 0.4585714286, 0.8640571429], [0.0629333333, 0.4736904762, 0.8554380952],
  37. [0.0722666667, 0.4886666667, 0.8467], [0.0779428571, 0.5039857143, 0.8383714286],
  38. [0.079347619, 0.5200238095, 0.8311809524], [0.0749428571, 0.5375428571, 0.8262714286],
  39. [0.0640571429, 0.5569857143, 0.8239571429], [0.0487714286, 0.5772238095, 0.8228285714],
  40. [0.0343428571, 0.5965809524, 0.819852381], [0.0265, 0.6137, 0.8135],
  41. [0.0238904762, 0.6286619048, 0.8037619048], [0.0230904762, 0.6417857143, 0.7912666667],
  42. [0.0227714286, 0.6534857143, 0.7767571429], [0.0266619048, 0.6641952381, 0.7607190476],
  43. [0.0383714286, 0.6742714286, 0.743552381], [0.0589714286, 0.6837571429, 0.7253857143],
  44. [0.0843, 0.6928333333, 0.7061666667], [0.1132952381, 0.7015, 0.6858571429],
  45. [0.1452714286, 0.7097571429, 0.6646285714], [0.1801333333, 0.7176571429, 0.6424333333],
  46. [0.2178285714, 0.7250428571, 0.6192619048], [0.2586428571, 0.7317142857, 0.5954285714],
  47. [0.3021714286, 0.7376047619, 0.5711857143], [0.3481142857, 0.7424333333, 0.5472666667],
  48. [0.3952571429, 0.7459, 0.5244428571], [0.4420095238, 0.7480809524, 0.5033142857],
  49. [0.4871238095, 0.7490619048, 0.4839761905], [0.5300285714, 0.7491142857, 0.4661142857],
  50. [0.5708571429, 0.7485190476, 0.4493904762], [0.609852381, 0.7473142857, 0.4336857143],
  51. [0.6473, 0.7456, 0.4188], [0.6834190476, 0.7434761905, 0.4044333333],
  52. [0.7184095238, 0.7411333333, 0.3904761905], [0.7524857143, 0.7384, 0.3768142857],
  53. [0.7858428571, 0.7355666667, 0.3632714286], [0.8185047619, 0.7327142857, 0.3497904762],
  54. [0.8506571429, 0.7299, 0.3360285714], [0.8824333333, 0.7274333333, 0.3217],
  55. [0.9139333333, 0.7257857143, 0.3062761905], [0.9449571429, 0.7261142857, 0.2886428571],
  56. [0.9738952381, 0.7313952381, 0.266647619], [0.9937714286, 0.7454571429, 0.240347619],
  57. [0.9990428571, 0.7653142857, 0.2164142857], [0.9955333333, 0.7860571429, 0.196652381],
  58. [0.988, 0.8066, 0.1793666667], [0.9788571429, 0.8271428571, 0.1633142857],
  59. [0.9697, 0.8481380952, 0.147452381], [0.9625857143, 0.8705142857, 0.1309],
  60. [0.9588714286, 0.8949, 0.1132428571], [0.9598238095, 0.9218333333, 0.0948380952],
  61. [0.9661, 0.9514428571, 0.0755333333], [0.9763, 0.9831, 0.0538],
  62. ]
  63. parula = ListedColormap(_parula_data, name='parula')
  65. # --- Paths ---
  66. source_data_dir = join(os.path.dirname(__file__), "..", "source_data", "Fig6")
  69. # ===========================
  70. # ==== Fitting & Plotting Helpers ====
  71. # ===========================
  73. def gaussian_with_baseline(x, amplitude, mean, stddev, baseline):
  74. return amplitude * np.exp(-((x - mean) ** 2) / (2 * stddev ** 2)) + baseline
  77. def gaussian_curve_fitting(x, y, constrained=True):
  78. df_tmp = pd.DataFrame({"x": x, "y": y})
  79. df_tmp = df_tmp.groupby("x").agg({"y": ["mean", "std", "sem"]}).reset_index()
  80. x_uniq = df_tmp["x"]
  81. y_mean = df_tmp["y"]["mean"]
  82. x_range = np.max(x_uniq) - np.min(x_uniq)
  83. initial_params = [np.max(y_mean) - np.min(y_mean), x_uniq[np.argmax(y_mean)], x_range / 4, np.min(y_mean)]
  84. bounds = ([0.0, np.min(x_uniq), 0.01, 0.0],
  85. [1.25 * np.max(y_mean), np.max(x_uniq), x_range * 2, np.max(y_mean)]) if constrained else None
  86. try:
  87. params, param_cov = curve_fit(gaussian_with_baseline, x, y, p0=initial_params, bounds=bounds)
  88. explained_variance = 1 - np.var(y - gaussian_with_baseline(x, *params)) / np.var(y)
  89. except RuntimeError:
  90. params, param_cov, explained_variance = None, None, None
  91. return {"params": params, "param_cov": param_cov, "explained_variance": explained_variance}
  94. def linear_regression_fitting(x, y):
  95. model = LinearRegression()
  96. model.fit(x.reshape(-1, 1), y)
  97. explained_variance = model.score(x.reshape(-1, 1), y)
  98. return {"params": {"slope": model.coef_[0], "intercept": model.intercept_},
  99. "param_cov": None, "explained_variance": explained_variance}
  102. def gaussian_process_regression(x_train, y_train, n_eval_points=100):
  103. length_scale = (x_train.max() - x_train.min()) / 10
  104. df_tmp = pd.DataFrame({"x": x_train, "y": y_train})
  105. noise_var = df_tmp.groupby("x").agg({"y": ['var']})["y"]["var"].mean()
  106. y_var = y_train.var()
  107. if np.isnan(noise_var):
  108. noise_var = y_var
  109. kernel = (ConstantKernel(y_var, (y_var * 1e-2, y_var * 1e2)) *
  110. RBF(length_scale=length_scale, length_scale_bounds=(1e-2, 1e2)) +
  111. WhiteKernel(noise_level=noise_var * 0.1, noise_level_bounds=(noise_var * 1e-2, noise_var * 1e1)))
  112. gpr = GaussianProcessRegressor(kernel=kernel, alpha=0.0, n_restarts_optimizer=10)
  113. gpr.fit(x_train.reshape(-1, 1), y_train.reshape(-1))
  114. explained_variance = 1 - np.var(y_train - gpr.predict(x_train.reshape(-1, 1))) / np.var(y_train)
  115. x_eval = np.linspace(x_train.min(), x_train.max(), n_eval_points)
  116. y_mean, y_std = gpr.predict(x_eval.reshape(-1, 1), return_std=True)
  117. return {"gpr": gpr, "explained_variance": explained_variance,
  118. "x_eval": x_eval, "y_mean": y_mean.reshape(-1), "y_std": y_std}
  121. def anova_test_df(df, x_col="lin_dist", y_col="pref_unit_resp"):
  122. try:
  123. model = ols(f'{y_col} ~ C({x_col})', data=df).fit()
  124. anova_table = sm.stats.anova_lm(model, typ=2)
  125. F_value = anova_table.loc[f'C({x_col})', 'F']
  126. p_value = anova_table.loc[f'C({x_col})', 'PR(>F)']
  127. return {"F_value": F_value, "p_value": p_value,
  128. "stats_str": f"F-val: {F_value:.2f} | p-val: {p_value:.1e}",
  129. "anova_table": anova_table, "error": None}
  130. except Exception as e:
  131. return {"F_value": np.nan, "p_value": np.nan, "stats_str": "", "anova_table": None, "error": e}
  134. def regression_combined_plot(x_train, y_train, gauss_fit_results, ols_fit_results, gpr_fit_results, anova_results=None, title_str="", ax=None):
  135. if ax is None:
  136. fig, ax = plt.subplots()
  137. else:
  138. fig = ax.figure
  139. ax.plot(x_train, y_train, 'o', label='data', alpha=0.5)
  140. tmp_df = pd.DataFrame({"x": x_train, "y": y_train}).groupby("x").agg({"y": ["mean", "std"]}).reset_index()
  141. ax.errorbar(tmp_df["x"], tmp_df["y"]["mean"], yerr=tmp_df["y"]["std"],
  142. label='mean ± std', marker='D', color="C1", markersize=5, capsize=5, alpha=0.5, linestyle="")
  143. x_eval = np.linspace(x_train.min(), x_train.max(), 100)
  144. gauss_params = gauss_fit_results["params"]
  145. ols_params = ols_fit_results["params"]
  146. gpr_kernel_str = str(gpr_fit_results["gpr"].kernel_).replace("length_scale", "len").replace("noise_level", "noise")
  147. if gauss_params is not None:
  148. ax.plot(x_eval, gaussian_with_baseline(x_eval, *gauss_params), label='Gaussian fit', color="k")
  149. ax.plot(x_eval, ols_params["slope"] * x_eval + ols_params["intercept"], label='OLS fit', color="magenta", linestyle="--")
  150. ax.plot(gpr_fit_results["x_eval"], gpr_fit_results["y_mean"], label='GPR mean', color="red")
  151. ax.fill_between(gpr_fit_results["x_eval"],
  152. gpr_fit_results["y_mean"] - gpr_fit_results["y_std"],
  153. gpr_fit_results["y_mean"] + gpr_fit_results["y_std"],
  154. alpha=0.2, label="GPR std", color="red")
  155. ax.legend()
  156. caption = f"{title_str}\n" if title_str else ""
  157. if gauss_params is not None:
  158. caption += f"Gauss: Ampl={gauss_params[0]:.2f} Mean={gauss_params[1]:.2f} Std={gauss_params[2]:.2f} Bsl={gauss_params[3]:.2f} [R2={gauss_fit_results['explained_variance']:.2f}]\n"
  159. caption += f"OLS: Slope={ols_params['slope']:.2f} Int={ols_params['intercept']:.2f} [R2={ols_fit_results['explained_variance']:.2f}]\n"
  160. caption += f"GPR: {gpr_kernel_str} [R2={gpr_fit_results['explained_variance']:.2f}]"
  161. if anova_results is not None:
  162. caption += f"\nANOVA: {anova_results['stats_str']}"
  163. ax.set_title(caption, fontsize=10)
  164. fig.tight_layout()
  165. return fig
  168. def plot_heatmap(grouped, space, ax, CLIM, annot=True, fmt=".1f"):
  169. space_data = grouped[grouped['space_name'] == space]
  170. pivot_table = space_data.pivot(index='eig_id', columns='lin_dist', values='pref_unit_resp').astype(float)
  171. if pivot_table.empty:
  172. return
  173. plt.sca(ax)
  174. sns.heatmap(pivot_table, annot=annot, fmt=fmt, cmap=parula,
  175. cbar_kws={'label': 'Preferred Unit Response'}, ax=ax, vmin=CLIM[0], vmax=CLIM[1])
  176. plt.title(f'Heatmap of Preferred Unit Response: {space} space')
  177. plt.xlabel('Linear Distance')
  178. plt.ylabel('Eigenvalue ID')
  179. plt.xticks(rotation=45, ha='right')
  180. plt.yticks(rotation=0)
  183. # ===========================
  184. # ==== Figure 6C: Example tuning curve ====
  185. # ===========================
  186. print("=== Figure 6C: Example tuning curve ===")
  187. sgtr_resp_df = pd.read_csv(join(source_data_dir, "Figure6C_src_B07092020_sgtr_resp_df.csv"))
  188. title_str = "B-07092020-003 | Pref Chan 5B"
  189. sgtr_resp_at_origin = sgtr_resp_df.query("lin_dist == 0.0")
  190. fig, ax = plt.subplots(nrows=1, ncols=1, figsize=(5, 5))
  191. space, eig_id = "class", 0
  192. sgtr_resp_per_axis = sgtr_resp_df.query("space_name == @space and eig_id == @eig_id")
  193. if 0.0 not in sgtr_resp_per_axis["lin_dist"].unique():
  194. sgtr_resp_per_axis = pd.concat([sgtr_resp_per_axis, sgtr_resp_at_origin])
  195. gauss_fit_results = gaussian_curve_fitting(sgtr_resp_per_axis["lin_dist"].values, sgtr_resp_per_axis["pref_unit_resp"].values)
  196. ols_fit_results = linear_regression_fitting(sgtr_resp_per_axis["lin_dist"].values, sgtr_resp_per_axis["pref_unit_resp"].values)
  197. gpr_fit_results = gaussian_process_regression(sgtr_resp_per_axis["lin_dist"].values, sgtr_resp_per_axis["pref_unit_resp"].values)
  198. anova_results = anova_test_df(sgtr_resp_per_axis)
  199. regression_combined_plot(sgtr_resp_per_axis["lin_dist"].values, sgtr_resp_per_axis["pref_unit_resp"].values,
  200. gauss_fit_results, ols_fit_results, gpr_fit_results,
  201. anova_results=anova_results, title_str=f"{title_str} | {space} {eig_id} axis", ax=ax)
  202. plt.xlabel("Linear Distance")
  203. plt.ylabel("Response (events/s)")
  204. plt.tight_layout()
  205. plt.show()
  207. # ===========================
  208. # ==== Figure 6D: Heatmap of tuning landscape ====
  209. # ===========================
  210. print("=== Figure 6D: Heatmap of tuning landscape ===")
  211. avgresp_df = pd.read_csv(join(source_data_dir, "Figure6D_src_prefchan_avgresp_df.csv"))
  212. ephysFN = avgresp_df.iloc[0]["ephysFN"]
  213. prefchan_str = avgresp_df.iloc[0]["prefchan_str"]
  214. CLIM = np.quantile(avgresp_df['pref_unit_resp'], [0.02, 0.98])
  215. figh, axs = plt.subplots(1, 2, figsize=(13, 6))
  216. for ax, space in zip(axs, ['class', 'noise']):
  217. plot_heatmap(avgresp_df, space, ax, CLIM)
  218. ax.set_title(f'{space} space')
  219. axs[-1].figure.axes[-1].set_ylabel("Response (events/s)")
  220. plt.suptitle(f'Preferred Unit Response\n{ephysFN} | Pref Channel {prefchan_str}')
  221. plt.tight_layout()
  222. plt.show()
  224. # Non-successful evolution example
  225. avgresp_df_fail = pd.read_csv(join(source_data_dir, "Figure6D_src_prefchan_avgresp_df_failevol.csv"))
  226. ephysFN = avgresp_df_fail.ephysFN.iloc[0]
  227. prefchan_str = avgresp_df_fail.prefchan_str.iloc[0]
  228. CLIM = np.quantile(avgresp_df_fail['pref_unit_resp'], [0.02, 0.98])
  229. figh, ax = plt.subplots(1, 1, figsize=(4.5, 4))
  230. plot_heatmap(avgresp_df_fail.query("eig_id in [0,1,2,3,6,9,13,21,30,60]"), "class", ax, CLIM, annot=False)
  231. ax.set_title('class space (failed evolution)')
  232. ax.figure.axes[-1].set_ylabel("Response (events/s)")
  233. plt.suptitle(f'Preferred Unit Response\n{ephysFN} | Pref Channel {prefchan_str}')
  234. plt.tight_layout()
  235. plt.show()
  237. # ===========================
  238. # ==== Figure 6E: Tuning shape & peak location ====
  239. # ===========================
  240. print("=== Figure 6E: Tuning shape type barplot ===")
  241. tuning_fitting_stats_table_sel = pd.read_csv(join(source_data_dir, "Figure6E_src_tuning_shape_fitting_stats_synopsis_selcolumn.csv"))
  242. filtered_data = tuning_fitting_stats_table_sel.query("anova_p_value < 0.01 and is_common_axis")
  243. melted_data = filtered_data.melt(
  244. id_vars='is_BigGAN_evol_success',
  245. value_vars=['gpr_y_is_bellshaped', 'gpr_y_is_monotonic', 'gpr_y_is_unimodal'],
  246. var_name='Metric', value_name='Value'
  247. )
  248. annotation_data = melted_data.groupby(['is_BigGAN_evol_success', 'Metric']).agg(
  249. True_Count=('Value', 'sum'), Total_Count=('Value', 'count'), Ratio=('Value', 'mean'),
  250. ).reset_index().sort_values(['is_BigGAN_evol_success', 'Metric'])
  252. plt.figure(figsize=(4.5, 6))
  253. ax = sns.barplot(data=melted_data, x='is_BigGAN_evol_success', y='Value', hue='Metric',
  254. order=["True", "False"], hue_order=["gpr_y_is_bellshaped", "gpr_y_is_monotonic"],
  255. errorbar=("ci", 95))
  256. plt.ylabel("Fraction of tuning axis")
  257. plt.xlabel("BigGAN evolution success")
  258. for p in ax.patches:
  259. height = p.get_height()
  260. row = annotation_data[annotation_data['Ratio'] == height]
  261. if row.empty:
  262. continue
  263. row = row.iloc[0]
  264. ax.annotate(f"{row['Ratio']:.2f}\n{int(row['True_Count'])}/{int(row['Total_Count'])}",
  265. (p.get_x() + p.get_width() / 2., p.get_height() / 2),
  266. ha='center', va='center', color="white", fontweight="bold")
  267. plt.suptitle("Tuning curve shape type as a function of BigGAN evolution success\n[signif axis, ANOVA p < 0.01]")
  268. plt.show()
  270. print("=== Figure 6E: Peak location distribution ===")
  271. tuning_stats_synopsis_df = pd.read_csv(join(source_data_dir, "Figure6E_src_tuning_stats_synopsis_selcolumn.csv"))
  272. fs = 10
  273. pval_threshold = 0.01
  274. tuning_stats_synopsis_df['sig'] = tuning_stats_synopsis_df['p_value'] < pval_threshold
  276. def plot_peak_location_panel(ax, df, title, show_ylabel=False):
  277. sns.countplot(data=df, x='max_resp_lin_dist_bin', stat='proportion', ax=ax, color='k')
  278. ax.set_title(f'{title} [N={len(df)}]', fontsize=fs)
  279. ax.set_xlabel('Peak Location on Axis', fontsize=fs)
  280. ax.set_ylabel('Fraction' if show_ylabel else '', fontsize=fs)
  281. counts = df['max_resp_lin_dist_bin'].value_counts().sort_index()
  282. N = counts.sum()
  283. props = counts / N
  284. ci_low, ci_high = proportion_confint(counts.values, N, alpha=0.05, method='wilson')
  285. bar_positions = [patch.get_x() + patch.get_width() / 2 for patch in ax.patches]
  286. ax.errorbar(bar_positions, props.values, yerr=[props.values - ci_low, ci_high - props.values],
  287. fmt='none', capsize=3, linewidth=1.5, color='red')
  288. for xpos, count in zip(bar_positions, counts.values):
  289. ax.text(xpos, 0, str(count), ha='center', va='bottom', fontsize=fs - 1, color='white')
  291. fig, ax = plt.subplots(1, 2, figsize=(6.5, 3.5), sharey=True)
  292. plot_peak_location_panel(ax[0], tuning_stats_synopsis_df.query("is_BigGAN_evol_success and sig"), "Successful", show_ylabel=True)
  293. plot_peak_location_panel(ax[1], tuning_stats_synopsis_df.query("not is_BigGAN_evol_success and sig"), "Non-successful")
  294. for axi in ax:
  295. for label in axi.get_xticklabels():
  296. label.set_rotation(45)
  297. label.set_ha('right')
  298. axi.tick_params(axis='x', labelsize=fs)
  299. fig.suptitle(f'Peak Response Location for successful vs non-successful evolution\n[Tuning Axis Signif. p < {pval_threshold}]', fontsize=fs)
  300. plt.tight_layout()
  301. plt.show()

Figure6_reproduction.py at commit 52e0cc7, under MIT · at the source

Overview

Authors: Binxu Wang1, Carlos R. Ponce2
ORCID iDs: Binxu Wang, Carlos R. Ponce
  1. Kempner Institute for the Study of Natural and Artificial Intelligence, Harvard University,Allston, MA USA
  2. Department of Neurobiology, Harvard Medical School,Boston, MA USA
Institutions: Harvard University (United States)
Journal: Nature neuroscience, volume 29, issue 4, pages 864-875
Dates: received 17 May 2024; accepted 9 January 2026; published online 10 March 2026; in print 2026
Type: Research article · Language: English
License: CC BY
Identifiers: DOI 10.1038/s41593-026-02207-1 · PMID 41807846 · PMCID PMC13061647 · OpenAlex W7134936292
Open access: hybrid, a free copy (OpenAlex)
Status: code verified
Categories: non-human primate (organism), systems (subfield)
Methods: Statistics, Connectivity, Machine learning, Single-unit activity, calcium imaging
Keywords: Object vision, Pattern vision
MeSH: Neurons*, Pattern Recognition, Visual*, Visual Cortex*, Animals, Macaca mulatta, Models, Neurological, Photic Stimulation, Visual Pathways (* major topic)
Topic: Face Recognition and Perception (Cognitive Neuroscience, Neuroscience), according to OpenAlex
Funding: U.S. Department of Health & Human Services | NIH | National Eye Institute (NEI) (DP2EY035176-01, P30EY012196-26); NSF (CAREER 2143077); David and Lucile Packard Foundation (David & Lucile Packard Foundation) (2020-71377)
Citations: cited by 2 papers (Europe PMC); 68 references in the paper

Abstract

Visual neurons respond to a vast range of images, from textures to objects, but the rules linking these responses remain unclear. Although tuning to simple features is well established in the primary visual cortex, this framework breaks down in higher areas, where neurons encode diverse and unpredictable features. To ask what features neurons prioritize, we used generative models (deep networks that synthesize new images from a learned latent space), allowing neurons in V1, V4 and the posterior inferotemporal cortex (PIT) to guide image synthesis through closed-loop optimization. We compared models that emphasize texture versus those that emphasize object structure. Although V1 and V4 aligned more strongly with texture-based spaces, many PIT neurons responded equally well to both types of optimized images, revealing a focus on shared local motifs rather than whole-object templates, and this alignment to objects emerged later in their response. These findings reveal coding principles across the ventral stream and clarify the limits of current vision models.

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

Repository

Its files are read in the Code ↔ Paper reader above, with 9 matches between paragraphs and lines of code.

Animadversio/Dynamic-Neuron-GAN-Alignment

License: MIT
State: the link answers, verified on 30 September 2026
Evidence: files inventoried
Commit: 52e0cc767155e05852edb8e9371f84d4c44ae596, 5 April 2026
Languages: Python (23), Jupyter (7)
Size: 34 files, 30 scripts
Software Heritage: not archived
Found in: “Code availability”
Holds: README, license file, environment (requirements.txt), 7 notebooks
Not found: CITATION.cff, tests, continuous integration, documentation
Tools: NumPy (23 files), Matplotlib (21 files), pandas (17 files), seaborn (16 files), SciPy (15 files), PyTorch (7 files), Pillow (3 files), statsmodels (3 files), scikit-image (2 files), scikit-learn (1 file)
Availability: 1 check, the latest on 30 September 2026: the link answers
  • 30 September 2026: the link answers
32 files
At the source: github.com/Animadversio/Dynamic-Neuron-GAN-Alignment

Code availability

The code used to analyze data in this paper is available on GitHub at https://github.com/Animadversio/Dynamic-Neuron-GAN-Alignment.

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

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;
  • 30 scripts, each with its path and the digest of its content;
  • 9 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

Data availability

Processed data for the analysis and reproduction of the figures can be downloaded at https://osf.io/pre96. The complete raw dataset used in this study is available upon request to C.R.P. Source data are provided with this paper.

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, 30 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 2 authors, 2 keywords, 8 MeSH terms, 3 funders, 48 references.

Cite

This paper

Wang, B., & Ponce, C. R. (2026). Neuronal tuning aligns dynamically with object and texture manifolds across the visual hierarchy. Nature neuroscience, 29(4), 864-875. https://doi.org/10.1038/s41593-026-02207-1

BibTeX

@article{wang2026neuronal,
author = {Wang, Binxu and Ponce, Carlos R.},
title = {{Neuronal tuning aligns dynamically with object and texture manifolds across the visual hierarchy}},
journal = {Nature neuroscience},
year = {2026},
month = mar,
volume = {29},
number = {4},
pages = {864--875},
publisher = {Nature Portfolio},
issn = {1097-6256},
doi = {10.1038/s41593-026-02207-1},
url = {https://doi.org/10.1038/s41593-026-02207-1},
pmid = {41807846},
pmcid = {PMC13061647}
}

RIS

TY - JOUR
AU - Wang, Binxu
AU - Ponce, Carlos R.
TI - Neuronal tuning aligns dynamically with object and texture manifolds across the visual hierarchy
T2 - Nature neuroscience
J2 - Nat Neurosci
PY - 2026
DA - 2026/03/10
VL - 29
IS - 4
SP - 864
EP - 875
SN - 1097-6256
PB - Nature Portfolio
DO - 10.1038/s41593-026-02207-1
UR - https://doi.org/10.1038/s41593-026-02207-1
LA - en
ER -

CSL-JSON

{
"id": "10.1038/s41593-026-02207-1",
"type": "article-journal",
"title": "Neuronal tuning aligns dynamically with object and texture manifolds across the visual hierarchy",
"container-title": "Nature neuroscience",
"author": [
{
"family": "Wang",
"given": "Binxu"
},
{
"family": "Ponce",
"given": "Carlos R."
}
],
"container-title-short": "Nat Neurosci",
"volume": "29",
"issue": "4",
"page": "864-875",
"DOI": "10.1038/s41593-026-02207-1",
"PMID": "41807846",
"PMCID": "PMC13061647",
"ISSN": "1097-6256",
"publisher": "Nature Portfolio",
"URL": "https://doi.org/10.1038/s41593-026-02207-1",
"language": "en",
"issued": {
"date-parts": [
[
2026,
3,
10
]
]
}
}

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/s41467-026-76106-1 [code]
Facial expression discrimination emerges from partially overlapping neural subspaces of detection and identity.
Journal: Nature communications
In common: seaborn, scikit-learn, pandas, 3 other tools, non-human primate, systems, 9 references
[2] doi:10.1038/s41467-026-76098-y [code]
A single computational objective can produce specialization of streams in visual cortex.
Journal: Nature communications
In common: scikit-image, Pillow, statsmodels, 7 other tools, 4 references
[3] doi:10.1038/s41467-026-74816-0 [code]
Hierarchical optimization predicts plasticity in the macaque inferior temporal cortex following object training.
Journal: Nature communications
In common: PyTorch, seaborn, scikit-learn, 4 other tools, non-human primate, 6 references
[4] doi:10.1038/s42003-026-10169-0 [code]
Shared representations in brains and models reveal a two-route cortical organization during scene perception.
Journal: Communications biology
In common: Pillow, statsmodels, PyTorch, 6 other tools, 4 references
[5] doi:10.1162/imag.a.1207 [code]
Investigating the temporal dynamics and modeling of mid-level feature representations in humans.
Journal: Imaging neuroscience (Cambridge, Mass.)
In common: Pillow, statsmodels, PyTorch, 6 other tools, 3 references
[6] doi:10.1523/jneurosci.0038-26.2026 [code]
Multidimensional Feature Tuning in Category Selective Areas of Human Visual Cortex.
Journal: The Journal of neuroscience : the official journal of the Society for Neuroscience
In common: Pillow, statsmodels, PyTorch, 6 other tools, systems, 2 references
[7] doi:10.3389/fncir.2026.1783892 [code]
Representations of facial features and surface quality in monkey area TE compared to neural network models.
Journal: Frontiers in neural circuits
In common: Pillow, PyTorch, NumPy, non-human primate, 5 references
[8] doi:10.1038/s41467-026-73938-9 [code]
Higher visual areas act like domain-general filters with strong selectivity and functional specialization.
Journal: Nature communications
In common: scikit-image, Pillow, PyTorch, 4 other tools, 3 references
[9] doi:10.1016/j.celrep.2026.117420 [code]
Neural population dynamics of direct electrical stimulation of neocortex.
Journal: Cell reports
In common: scikit-image, Pillow, statsmodels, 6 other tools, systems, 1 reference
[10] doi:10.1371/journal.pcbi.1013138 [code]
Hierarchical recurrent temporal prediction as a model of the mammalian dorsal visual pathway.
Journal: PLoS computational biology
In common: scikit-image, statsmodels, PyTorch, 5 other tools, systems, 2 references

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.