Virtual Brain and Digital Twins in Neurogenetics: From Multimodal Patient Data to Genomically Informed, Clinically Actionable Models

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This review proposes a genetics-centered framework for virtual brain modeling and digital twin construction in neurogenetic disorders, organized around a three-tier evidence ladder (L1–L3) that maps genetic information onto model components at increasing levels of patient specificity. The manuscript covers relevant data streams, provides disease exemplars (genetic FTD, Huntington's disease, HSP), and includes a useful discussion of validation and reporting requirements. The writing is cautious and appropriately scoped, and the L1–L3 structure provides a clear organizing logic. That said, I think the manuscript would benefit from better contextualization of the proposed framework against prior work that has already implemented similar ideas, and from some adjustments to distinguish the author's novel contributions from the review of existing literature. My specific comments are below.

1. The L1–L3 evidence ladder is the author's own formalization, and the core of Section 4 reads more as a perspective-style proposal than a systematic review. I think this is fine, but prior work that has already implemented key elements of L1 and L2 should be discussed to contextualize the contribution. For example, Demirtaş et al. (Neuron, 2019) and Fulcher et al. (Science Advances, 2021) incorporated transcriptomic and hierarchy-derived heterogeneity into whole-brain biophysical models, and Freeze et al. (2018, 2019) integrated disease-gene expression into network spreading models for Parkinson's disease. I would encourage the authors to discuss how their proposed framework relates to and builds upon these prior implementations.

2. Given that this framework proposes clinical use of individual genomic data, a brief discussion of ethical considerations, particularly equity implications of ancestry-dependent PRS performance and the risk of over-interpreting unvalidated model outputs, would strengthen the manuscript.

3. Section 4.2 appears twice, once for L2 (line 175) and once for L3 (line 204). The latter should be renumbered to 4.3.

4. Line 25: There is a double period (“diseases..”) at the end of the abstract. Similarly, line 480 has the same issue.

5. Table 1 is informative but difficult to read in its current format.

6. Several abbreviations (e.g., CNV, EEG, MEG) are used without being defined at first mention.

7. The Author Contributions section lists a single author for all roles including “Software” and “Data curation,” but this is a review article with no original software or data? 

Author Response

I sincerely thank the reviewers for their careful reading of the manuscript and for their thoughtful and constructive comments. Their observations were highly valuable and prompted substantial revisions that improved the manuscript in terms of clarity, conceptual positioning, methodological precision, and overall interpretability.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript proposes a genetics-centered framework for constructing virtual brain models and medical digital twins in the context of neurogenetic disorders. The author introduces a three-tiered evidence ladder (L1-L3) that progressively integrates transcriptomic atlases, genotype-linked network phenotypes, and patient-specific exome/CNV data into whole-brain biophysical simulations. The overarching goal is to advance the field beyond the use of genetics as a mere diagnostic label, positioning it instead as a constructive input to model design.
The topic is timely, conceptually well-organized, and clinically motivated. The manuscript covers a broad territory, from imaging transcriptomics and connectome modeling to polygenic risk scores and digital twin governance, with appropriate intellectual ambition. Nevertheless, the review has notable shortcomings that currently limit its impact. Several sections lack methodological depth, the literature coverage is selective in certain areas, and the proposed L3 roadmap, while aspirationally described, requires a clearer articulation of existing barriers and concrete validation strategies. The following points detail specific concerns across major and minor categories.
The manuscript's core contribution, positioning genetics as a 'constructive input' rather than a diagnostic label, is intellectually compelling but incompletely differentiated from existing literature. The review does not clearly articulate what is genuinely novel in the proposed L1-L3 ladder versus prior frameworks that have similarly discussed multi-level genetic integration in computational neurology. A dedicated paragraph contrasting the proposed approach with at least two or three established frameworks would substantially strengthen the justification for the review's conceptual contribution.
Furthermore, the relationship between 'virtual brain' and 'digital twin' is addressed in Section 2 but remains conceptually blurry throughout the manuscript. The author defines a virtual brain twin as 'a domain-specific instantiation of a digital twin focused on brain dynamics and network physiology,' yet the manuscript inconsistently uses these terms in later sections. A clear terminological convention, established early and enforced consistently, would improve readability and scholarly precision.
The L1-L3 ladder is the intellectual centerpiece of the manuscript, yet its operationalization is uneven. L1 (atlas-based transcriptomic priors) is well-grounded with appropriate citations to imaging transcriptomics best practices and the Allen Human Brain Atlas. L2 (disease-linked network phenotypes) is adequately illustrated through the FTD/ALS/Huntington's disease examples. However, L3 (patient-specific exome/CNV integration) is described almost entirely in aspirational terms. The author acknowledges that L3 'remains a roadmap until mapping rules are prospectively validated,' yet no concrete metrics, milestones, or preliminary data are cited to support the feasibility of this transition. 
The manuscript acknowledges that data streams are the 'practical bottleneck' for any brain digital twin (Section 3), yet the discussion of data acquisition technologies is limited to clinical and conventional research-grade modalities (structural MRI, EEG/MEG, wearables). An important and growing literature on next-generation biosensing platforms is entirely absent from this review. For instance, recent advances in ultra-sensitive, multimodal on-chip biosensors, capable of detecting nucleic acids at attomolar concentrations via simultaneous electrochemical, field-effect transistor (FET), and optical (SERS-based) readout, represent a potentially transformative avenue for the type of molecular data integration envisioned at L3.
Specifically, nano-corrugated graphene (NCGr)-based devices reported by Nik Zulkarnine et al. (2025, Device, 3, 100572) demonstrate that a single on-chip platform can support three distinct detection modes, electrolyte-gated FET, electrochemical, and SERS, with DNA detection sensitivity down to 1 aM. The study further demonstrates that charge-transfer mechanisms dominate across all three modalities, enabling reliable and reproducible biosensing under near-physiological conditions. This class of platform is directly relevant to the challenge of generating high-resolution molecular phenotyping data at the patient level, which is precisely what L3 of the proposed framework requires. The manuscript would benefit from explicitly acknowledging this technological direction and discussing how ultra-sensitive, multiplexed molecular biosensors could provide the genomic and proteomic data streams needed to populate and validate patient-specific priors within a digital twin pipeline.
More broadly, a brief subsection discussing emerging data acquisition technologies, beyond conventional neuroimaging and electrophysiology, would align well with the manuscript's stated ambition of integrating multimodal data into clinically actionable models. The practical utility of a digital twin framework depends as much on the richness of its input data as on the sophistication of its computational architecture.
Section 5 presents the Virtual Epileptic Patient (VEP) and Bayesian VEP (BVEP) as a 'proof of feasibility' for the broader neurogenetics framework. While this choice is well-justified given the maturity of the epilepsy modeling literature and the availability of SEEG-based ground truth, the transferability argument is underdeveloped. The author briefly notes that 'translation to neurogenetic disorders should be framed as methodological transfer rather than direct extrapolation,' but does not systematically address the key structural differences between epilepsy and the target neurogenetic diseases discussed in Section 6.
Section 7.4 discusses validation strategies under rare-disease constraints but remains at a high level of generality. The reviewer recognizes the inherent difficulty of validation in rare disease settings; however, the current discussion does not sufficiently distinguish between what is feasible now with available cohort infrastructure (e.g., GENFI for genetic FTD, TRACK-HD/ENROLL-HD for Huntington's) versus what requires novel data collection. 
Section 7.5 briefly addresses clinical implementation requirements, including governance and audit trails. However, the manuscript does not meaningfully engage with the regulatory pathway for deploying a genetics-centered digital twin in clinical practice, nor with equity concerns arising from the differential performance of polygenic risk scores (PRS) across ancestries, an issue the author correctly identifies in Section 4.3 but does not follow through in the clinical translation discussion. Given the growing literature on algorithmic bias in genomic medicine, a dedicated paragraph addressing equity, fairness, and regulatory readiness would substantially strengthen the manuscript's relevance for clinical audiences.
The abstract is technically sound and well-structured. However, the final sentence ('The realistic near-term impact is not deterministic prediction, but better stratification, trajectory forecasting, and scenario-based decision support in rare neurogenetic diseases') would benefit from being more specific about the time horizon and the conditions under which this impact is achievable.
The manuscript contains two figures (Figures 1 and 2). Both are schematic in nature and convey the overall framework effectively. However, Figure 2 (Genotype-to-Network Translation) lacks sufficient detail in its legend. Specifically, the depiction of 'connectome-constrained' modeling in the Huntington's disease panel is not self-explanatory. The legend should clarify what the network visualization represents (e.g., structural connectome, functional coupling, spread model output) and what the colormap encodes.
Table 1 is a valuable synthesis of the genetics-to-model-component mapping. The reviewer suggests adding a column for 'Current Readiness Level' (e.g., Proof-of-concept / Emerging / Established) to make the translational status of each row explicit. This would complement the L1-L3 labeling and provide readers with an at-a-glance summary of the field's maturity.
The term 'digital phenotyping' is used in two distinct senses in the manuscript: (i) wearable/remote monitoring-based assessment (Section 3.1) and (ii) computational phenotyping from multimodal data (abstract, keywords). This dual usage may confuse readers. The author should clearly distinguish between these two meanings, ideally by adopting the term 'remote digital phenotyping' for the wearable-based sense and reserving 'digital phenotyping' for the broader computational sense, or vice versa, as long as the choice is explicit.
 
The manuscript addresses an important and underexplored intersection of computational neuroscience and neurogenetics. The conceptual framework is sound, and the tiered evidence ladder represents a useful organizational contribution to the field. However, the manuscript currently falls short in its operationalization of L3, its engagement with emerging data acquisition technologies (including next-generation biosensing), the depth of its validation guidance, and its treatment of equity and regulatory considerations. The reviewer recommends Major Revision, with specific attention to the points raised above, particularly the biosensing integration argument, the L3 feasibility discussion, and the validation design guidance.
 
A revised version that addresses these concerns would represent a substantive and timely contribution to the literature on personalized computational medicine in rare neurological diseases.

Author Response

I sincerely thank the reviewers for their careful reading of the manuscript and for their thoughtful and constructive comments. Their observations were highly valuable and prompted substantial revisions that improved the manuscript in terms of clarity, conceptual positioning, methodological precision, and overall interpretability.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for your thorough revision. The authors have addressed all of my comments satisfactorily, and I am happy to recommend this manuscript for acceptance.

Reviewer 2 Report

Comments and Suggestions for Authors

It can be published as it is now.