Uncropped Film Scans and Analysis Data Addressing PubPeer Query on JNC 2011 vs JNC 2009_Socodato et al.

This dataset addresses allegations related to cross‑paper similarities between Paper 1 (2009, J Neurochem) and Paper 2 (2011, J Neurochem): (i) Fig. 2d (AKT) vs Fig. 4e (AKT), and (ii) Fig. 3b/5a (pCREB & AKT) vs Fig. 4c. We re‑analyzed the uncropped films with a lane‑agnostic, ECC‑aligned computer‑vision pipeline. The PubPeer pages showing the annotated panels are attached for context. Concern 1 — Paper 1, Fig. 2d (AKT) vs Paper 2, Fig. 4e (AKT) Lane mapping tested: 2d‑L1 ↔ 4e‑L2, 2d‑L2 ↔ 4e‑L3, 2d‑L3 ↔ 4e‑L4. Defaults: full‑height search; band ≈ 8% of image height; ECC alignment; uncropped films. Results (per‑lane metrics, ECC‑aligned): 2d‑L1 ↔ 4e‑L2: NCC −0.170, SSIM 0.579, Edge‑Hamming 0.094, Profile_r 0.648, Residual_corr −0.052. 2d‑L2 ↔ 4e‑L3: NCC 0.909, SSIM 0.983, Edge‑Hamming 0.152, Profile_r 0.729, Residual_corr −0.026. 2d‑L3 ↔ 4e‑L4: NCC 0.276, SSIM 0.982, Edge‑Hamming 0.191, Profile_r 0.427, Residual_corr −0.092. (ORB inlier ratio ≈ 0 for all three.) Interpretation. One pair (2d‑L2 ↔ 4e‑L3) shows high structural similarity (NCC≈0.91, SSIM≈0.98), but fails geometric/edge criteria: Edge‑Hamming remains high (0.152), ORB inliers ≈ 0, and residual correlations are negative. In our calibrated framework, true re‑use requires concordance across structure + edge + local geometry; here, the signal does not reach a multi‑metric duplication signature. We therefore do not find evidence that AKT bands were re‑used between Fig. 2d (2009) and Fig. 4e (2011). Concern 2 — “Very similar” pCREB & AKT between Paper 1 (Fig. 3b, Fig. 5a) and Paper 2 (Fig. 4c) Protocol: strictly protein‑matched comparisons; full‑height search; band ≈ 8%H; ECC alignment; uncropped films. AKT (3b lanes 2 & 3 → 4c all lanes, best hits): Target L2 best at 4c‑L1 — NCC 0.239, SSIM 0.874, Edge‑Hamming 0.234. Target L3 best at 4c‑L5 — NCC 0.277, SSIM 0.765, Edge‑Hamming 0.288. (ORB inlier ratio 0.0 in both best pairs.) pCREB (3b lanes 2 & 3 → 4c and 5a, best hits): 3b → 4c: L2 best NCC 0.330, SSIM 0.846; L3 best NCC 0.276, SSIM 0.943. 3b → 5a: L2 best NCC 0.577, SSIM 0.931; L3 best NCC 0.681, SSIM 0.914. Calibration vs. non‑match baselines (percentiles). For the AKT 3b→4c screen, the “best” hits fall inside the negative distributions: e.g., for 4c‑AKT target L2, NCC and SSIM sit at the 100th percentile of negatives, Edge‑Hamming ≈ 89th, Profile_r ≈ 100th; for target L3, SSIM ≈ 33rd and Edge‑Hamming ≈ 33rd percentile. (See percentiles.csv and best_pairs.csv.) Interpretation. The closest cross‑paper matches remain definitively non‑matches by calibrated criteria: structural similarity alone is insufficient without edge/geometry concordance, and all “best” pairs lie within their negative baselines. These datasets are independent films that share some gross morphology but diverge in micro‑texture and edge geometry. What the metrics mean NCC (−1…+1): normalized cross‑correlation after ECC; +1 identical. SSIM (0…1): structural similarity; values ~0.98 can occur without geometric agreement. Edge‑Hamming (0 identical → higher = different): Hamming distance of binarized edge maps; small values indicate edge‑shape identity. Profile_r (0…1): correlation of vertical intensity profiles; insensitive to micro‑texture differences. Residual_corr (−1…+1): correlation of difference images; values near 0/negative argue against duplication. ORB inlier ratio (0…1): fraction of keypoint matches surviving RANSAC; ≈0 indicates no coherent geometric mapping.