-
The central scientific objective (linking BMPAA to cognitive resilience) is not tested because behavioral feature extraction failed due to a systematic parsing error; consequently, none of the planned brain–behavior models described in Sec. 2.5 are executed (Secs. 2.2, 2.5, 3.2.1, 3.4, 4.2–4.4). As written, several sections still frame the work as if resilience were evaluated, which overstates what the current results can support.
Recommendation: Choose one of two paths and revise the manuscript consistently: (i) Fix the parsing/extraction pipeline, regenerate behavioral variables, and run the pre-specified models in Sec. 2.5 with full reporting ($N$ per model, effect sizes, uncertainty, diagnostics), or (ii) if recovery is not feasible, explicitly reframe as a methods/feasibility paper. For option (ii), revise the Abstract, Introduction, Sec. 2.5 (tense and positioning), Results (Sec. 3.2.1), and Conclusions (Secs. 4.2–4.4) to clearly state that cognition/resilience analyses are future work and remove/down-weight claims implying demonstrated cognitive relevance.
-
Imaging preprocessing, tensor fitting, ROI extraction, and atlas definition are under-specified, limiting reproducibility and interpretability of MD (Secs. 2.3.1–2.3.2). Key missing elements include: motion/eddy-current and distortion correction, tensor fitting method, registration/normalization to atlas/template space, smoothing (if any), QC/exclusion criteria, handling of partial-volume/CSF contamination, and the provenance/validation of the 24-ROI bat atlas (including the full ROI list). The presence of truncated/zero-filled path-like strings further obscures the pipeline.
Recommendation: Expand Sec. 2.3 into a stepwise, reproducible description: software and versions; preprocessing steps and parameters (e.g., eddy/motion, susceptibility correction if applicable); tensor fitting approach; template/atlas registration procedure; ROI summarization method (mean/median, voxel weighting); and QC metrics/thresholds. Add a table/appendix listing all 24 ROI labels with brief anatomical descriptions and atlas source/adaptation details for Rousettus. Replace internal absolute paths in Secs. 2.1.2/2.3.1/2.4.1 with human-readable placeholders and clarify whether any code/data will be shared.
-
Cohort reporting is inconsistent ($N=29$ vs $N=30$; sex/origin counts differ across sections), and these discrepancies propagate into the stated MD matrix dimension ($29\times24$ vs $30\times24$) used for PCA and regressions (Secs. 2.1.3, 3.1, 3.3.1; Table 1). This undermines confidence in curation and in all downstream quantitative results.
Recommendation: Reconcile the analytical cohort definitively: provide a single subject-flow summary (initial $N$, exclusions by reason, final $N$ per modality/analysis). Update Sec. 2.1.3, Sec. 3.1, Table 1, and figure captions to match exactly (including DNAmAge mean/SD). Explicitly state the final $N$ used for (i) MD matrix construction, (ii) PCA fitting, and (iii) each regression.
-
PCA component retention and stability are insufficiently justified given the small sample ($N \approx 30$) relative to 24 ROIs, and the reliance on Kaiser ($>1$ eigenvalue) plus a variance threshold may over-retain components (Secs. 2.4.2, 3.3.1). Without stability checks, it is unclear whether PC3’s age association (and thus BMPAA on that component) is robust to sampling variation.
Recommendation: Add robustness analyses for PCA (Secs. 2.4.2, 3.3.1): bootstrap or split-half resampling to quantify stability of eigenvalues and loadings (e.g., loading congruence/similarity), and include a parallel analysis (Horn) and/or scree-elbow justification. Report sensitivity of the key findings (PC3$\sim$DNAmAge; PC1$\sim$sex; PC2$\sim$origin) to retaining different numbers of PCs (e.g., 4–8). Frame conclusions explicitly as exploratory if stability is limited.
-
Regression models linking PC scores to DNAmAge/sex/origin and defining BMPAA are under-specified and create avoidable inferential risk: the equation in Sec. 2.4.3 omits Origin_colony despite text claiming inclusion; coding for categorical predictors is not stated; standardized $\beta$s are not reproducible from the Methods; and multiple comparisons across PCs$\times$predictors are not addressed (Secs. 2.4.3, 3.3.2). Visual diagnostics are presented, but quantitative influence/leverage assessment is not reported despite small $N$.
Recommendation: In Sec. 2.4.3, explicitly specify the full model used for each PC (e.g., $\mathrm{PC}_k \sim \mathrm{DNAmAge} + \mathrm{Sex} + \mathrm{Origin\_colony}$), including intercept, error term, and the coding/reference levels for Sex and Origin. State exactly what was standardized (PC scores and/or DNAmAge) to produce “standardized $\beta$”. In Sec. 3.3.2, report coefficient CIs and model fit ($R^2$; ideally partial $R^2$), and correct or control for multiplicity (e.g., FDR across 6 PCs for each predictor family, or a clearly stated exploratory stance). Add influence diagnostics (e.g., Cook’s distance/leverage summary) and note whether any results depend on single points.
-
Interpretability of PCs and of the term “age acceleration” is currently too thin. PC3 is labeled “age-related” largely because it correlates with DNAmAge, but the manuscript does not show which ROIs drive PC1–PC3 nor how the sign/direction of the component maps to higher/lower MD in anatomically meaningful systems (Secs. 3.3.1–3.3.2, 3.4, 4.3–4.4). Additionally, interpreting residual sign as “more aged” is only justified for components with a defined aging direction and after resolving the arbitrary PC sign; it is not appropriate for PCs primarily associated with sex/origin (Sec. 2.4.3).
Recommendation: Provide loading tables/plots for at least PC1–PC3 (top positive/negative ROI contributors) in Sec. 3.3.1–3.3.2 (main text or supplement), and interpret these patterns anatomically with relevant diffusion-aging context. Explicitly define the aging direction for the age-associated component (e.g., fix PC sign such that higher PC3 corresponds to higher predicted DNAmAge-associated MD pattern expression). Restrict “age acceleration” language to the age-associated component(s) and describe residuals for other PCs as “covariate-adjusted component expression,” not aging.
-
Sex and colony associations (PC1/PC2) may reflect technical/batch confounds (scanner/session differences, scan date, motion/SNR differences, processing batches) rather than biology, but technical covariates and potential confounding with age are not examined (Sec. 3.3.2; cohort description in Sec. 3.1).
Recommendation: Report available technical covariates (scan date/session, protocol changes, motion metrics, SNR/QC summaries) and test whether they correlate with sex/colony and/or with PCs. At minimum, add DNAmAge distributions by sex and by colony (and consider including chronological age if available) to assess confounding. If technical covariates exist, include them in sensitivity regressions or discuss explicitly as alternative explanations in Sec. 4.4.