Analytics

We noticed general trends in the analytical strategies used in the articles discussed here. In studies that adopt frameworks like a traditional GWAS (e.g., X. Li et al., Reference Li, Jiang, Zhao, Yang, Lu, Zhang, Li and Wang2024; Yu et al., Reference Yu, Liu, Wu, Becker, Zhang, Fan, Peng, Kuang, Kang, Dong, Zhao, Schumann, Feng, Sahakian, Robbins, Palaniyappan and Zhang2023), the analytical pipeline first consists of a method to find associations between the genotype and a trait-affected imaging phenotype (e.g., differences in cortical measures between affected and unaffected groups). The significantly associated genetic markers are then annotated to find corresponding relevant genetic loci or genes. The majority of the studies employed a microarray approach for genotyping, likely due to its cost-effectiveness and the reliability of the genotypes produced. These microarrays typically featured a marker density ranging from 300,000 to 600,000 markers. To enhance the resolution of genetic data, imputation was frequently utilized rather than “liftover” techniques that convert genetic coordinates from one assembly version to another. This imputation process significantly increases the number of markers, expanding datasets from hundreds of thousands to approximately 8 million markers (e.g., (Ge et al., Reference Ge, Wu, Zhang, Chen, Zhang, Kang, Deng, Ou, He, Zhao, Kuo, Ma, Banaschewski, Barker, Bokde, Desrivières, Flor, Grigis, Garavan and Yu2024).

A major limitation of GWAS-type studies is their reliance on large sample sizes to detect small effect sizes, making them less effective for capturing rare variants with potentially larger effects. While these studies are useful for identifying common genetic markers associated with traits, they often fall short of providing functional insights or mechanistic explanations of how these markers influence brain structure and function. Moreover, the significant computational demands of handling imputed datasets and the complexity of controlling for population stratification and other confounding variables pose challenges. As a result, complementary approaches, like enrichment analysis using resources like the STRING database (see Box 4), are often needed to gain a more comprehensive understanding of the genetic underpinnings of neurodevelopmental disorders and/or developmental psychopathology phenotypes.

Enrichment analysis is often conducted to determine if the collectively identified set of genes or loci are overrepresented in relation to certain biological functions, or pathways, or if the identified genes have been previously associated with similar traits. In some cases, enrichment analysis compares the gene groups flagged by a GWAS (or a similar approach for identifying associated markers) with pre-defined gene sets that correspond to gene ontology (GO) terms, pathways, or interaction networks and determines if the experimentally selected genes are statistically overrepresented in these pre-defined sets (e.g., Yu et al., Reference Yu, Liu, Wu, Becker, Zhang, Fan, Peng, Kuang, Kang, Dong, Zhao, Schumann, Feng, Sahakian, Robbins, Palaniyappan and Zhang2023).

Another limitation of GWAS in developmental research is that, despite the many significant SNP associations with neuroimaging traits, these identified variants typically exert only minor influences and represent just a small portion of all truly associated variants. As a result, their ability to predict outcomes is limited. To address this limitation, researchers have incorporated PRS to aggregate the many small effects of associated common variants (Dudbridge, Reference Dudbridge2013; Palla & Dudbridge, Reference Palla and Dudbridge2015). Studies incorporating PRS combine the effects of variants across the entire genome to achieve several goals: 1) estimate heritability, 2) identify genetic overlap between different traits, and 3) predict phenotypes based on an individual’s genetic profile. It appears that, with the exception of educational attainment, there is a limited number of these scores that are used regularly. Notably, for developmental disorders, a limited number of specific studies are commonly referenced, such as for ASD (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won, Pallesen, Agerbo, Andreassen, Anney, Awashti, Belliveau, Bettella, Buxbaum, Bybjerg-Grauholm, Bækvad-Hansen, Cerrato, Chambert, Christensen and Børglum2019) and ADHD (Demontis et al., Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo, Baldursson, Belliveau, Bybjerg-Grauholm, Bækvad-Hansen, Cerrato, Chambert, Churchhouse, Dumont, Eriksson, Gandal, Goldstein, Grasby, Grove and Neale2019). Such a limited pool of PRS sources suggests a need for a way to calculate PRS in an ongoing fashion, constantly diversifying or stratifying the contributing samples.

Many studies in our reviewed set used genetic components represented by DNA-sequencing data (e.g., Chilosi et al., Reference Chilosi, Podda, Ricca, Comparini, Franchi, Fiori, Pasquariello, Casalini, Cipriani and Santorelli2022), RNA-sequencing data (e.g., Ardesch et al., Reference Ardesch, Libedinsky, Scholtens, Wei and van den Heuvel2023), or a combination of RNA-seq and proteomics (e.g., (Hale et al., Reference Hale, Bastarache, Morales, Wellons, Limbrick and Gamazon2021) or RNA-seq and DNA methylation data (e.g., Zheng et al., Reference Zheng, Womer, Tang, Guo, Zhang, Tang and Wang2024). In cases where DNA-sequencing data serve as the genetic component, genomic analyses tended to precede imaging analysis. Chilosi and colleagues (2022) first identified copy number variations (CNVs) associated with the neurodevelopmental disorder in question (childhood apraxia of speech), and after that, they used the MRI data to identify any associated structural brain abnormalities. In cases where RNA-sequencing served as the genomic component, imaging data analysis often preceded genomic analysis. For example, Jabbi and colleagues (2020) determined regions of interest from MRI images and then analyzed RNA data (gene expression levels) in these specific regions. Depending on the study, authors can use transcriptomic data from open sources such as AHBA (Ardesch et al., Reference Ardesch, Libedinsky, Scholtens, Wei and van den Heuvel2023) to build spatial expression profiles across the region of interest. In studies that implement multi-omics approaches (e.g., Hale et al., Reference Hale, Bastarache, Morales, Wellons, Limbrick and Gamazon2021; Zheng et al., Reference Zheng, Womer, Tang, Guo, Zhang, Tang and Wang2024), researchers may first combine genetics, proteomic, and epigenetics data into the common ‘genetic component’ and use various databases as references to search for a specific disease(condition)-associated variants as potential markers. After that, MRI data for relevant phenotypes are being used to verify the findings. GMV data can also serve as one of the parameters in such multicomponent studies (Jabbi et al., Reference Jabbi, Arasappan, Eickhoff, Strakowski, Nemeroff and Hofmann2020; Zheng et al., Reference Zheng, Womer, Tang, Guo, Zhang, Tang and Wang2024).

Diagnoses

There appears to be a narrow breadth of developmental disorders examined from an imaging genetics perspective. Most of the synthesized articles focused on participants with ASD and ADHD. There are also developmental disorders, such as SCHZ, mood disorders, or broadly defined developmental psychopathology phenotypes such as internalizing and externalizing. The number of studies focusing on a particular disorder or related phenotypes seems to be related to the availability of publicly accessible imaging genetic data. As the needed databases are impossible (or near impossible) to secure at a single site, it is important that the field adopts common best practices in the methodology of data acquisition and processing as quickly as possible to ensure that the data are mergeable and findings are robust, replicable, and generalizable across different neurodevelopmental disorders and developmental psychopathology phenotypes.

Integration of multimodal data

As new data become available, imaging genetics studies increasingly rely on multimodal neuroimaging data, combining structural MRI, fMRI, EEG, and other techniques to better characterize a given disorder. While this approach offers a more nuanced understanding of how genetic factors influence the brain, it also introduces a new challenge involving scalability, interpretability, and generalization of complex association models for integrating heterogeneous data types (Biessmann et al., Reference Biessmann, Plis, Meinecke, Eichele and Müller2011). Some of the typical hurdles include incommensurability of different datasets, e.g., brain structure, activity, clinical scores, gene expression levels involved in the study, spatial and dimensional heterogeneity of neuroimaging data caused by the varying spatial resolution limit of each of these modalities, or the data types that range from, e.g., time series data from fMRI to tensors of diffusion-weighted imaging (Lorenzi et al., Reference Lorenzi, Deprez, Balelli, Aguila, Altmann and Colliot2023). Others include heterogeneous artifacts and measurement uncertainty resulting from varied acquisition and processing routines, missing data, and the interpretability of the obtained results. In recent years, various methods have been developed to address these issues, some of which include integrating sMRI and diffusion MRI (Liu & Calhoun, Reference Liu and Calhoun2014), EEG and fMRI (Rastogi et al., Reference Rastogi, Lee and Salamon2008), and sMRI and fMRI (Mueller et al., Reference Mueller, Keeser, Samson, Kirsch, Blautzik, Grothe, Erat, Hegenloh, Coates, Reiser, Hennig-Fast, Meindl and Draganski2013). Moreover, various mathematical frameworks like Bayesian latent variable models (Lorenzi et al., Reference Lorenzi, Deprez, Balelli, Aguila, Altmann and Colliot2023) and mediation model framework (Zhao et al., Reference Zhao, Li and Caffo2021) have also been developed for the integration of multimodal informational variables derived from these datasets.

A methodological hurdle associated with the growing amount of high-dimensional data inherent to both genetic and neuroimaging research is the computational infrastructure required for analysis and storage. The technological advancements in high-throughput microarray genotyping platforms allow more affordable and hypothesis-free interrogation of millions of genetic variants across the genome (Uffelmann et al., Reference Uffelmann, Huang, Munung, de Vries, Okada, Martin, Martin, Lappalainen and Posthuma2021). Similarly, the possibility of collecting multimodal neuroimaging data concurrently or the availability of open access databases have increased significantly over the last decades (Alexander et al., Reference Alexander, Escalera, Ai, Andreotti, Febre, Mangone, Vega-Potler, Langer, Alexander, Kovacs, Litke, O’Hagan, Andersen, Bronstein, Bui, Bushey, Butler, Castagna, Camacho and Milham2017; Jernigan et al., Reference Jernigan, Brown, Hagler, Akshoomoff, Bartsch, Newman, Thompson, Bloss, Murray, Schork, Kennedy, Kuperman, McCabe, Chung, Libiger, Maddox, Casey, Chang, Ernst and Dale2016). However, the amount of storage and computing power (e.g., high-performance computing, HPC, clusters) needed to analyze these datasets are not prioritized or available to a significant fraction of researchers. In the United States, there has been a concerted effort to increase the availability of HPC infrastructure through various institutions and government initiatives like the National Science Foundation (NSF) - Extreme Science and Engineering Discovery Environment (XSEDE) program, Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS), or Harnessing the Data Revolution (HDR) initiative, but the challenges related to demand, funding, technical expertise, and geographical disparities persist.

Sample size, reproducibility, and the challenge of small effect sizes

Reproducibility is a ubiquitous problem in scientific research, and imaging genetics is no exception. Basically, the reproducibility crisis describes the situation when published research results are too often not replicable, which leads to the slowdown of scientific progress in the field (Kelly & Hoptman, Reference Kelly and Hoptman2022). Avinun and colleagues (Reference Avinun, Nevo, Knodt, Elliott and Hariri2018) note that major contributing factors to the reproducibility crisis in imaging genetics include small effect sizes, heterogeneity between samples, and differences in analytical strategies. These issues are closely related to the issue of sample size and a lack of well-defined guidelines for imaging genetic research (i.e., which imaging modalities are most appropriate to interrogate which developmental outcome). Though, practically, a major limitation to choosing which imaging modality is most appropriate to study a given developmental population lies in which is best suited to acquire good data from them (e.g., small children may be more amenable to an EEG versus an MRI). This decision also depends on the phenotype and research question. Since EEG and MRI are not interchangeable, the choice will partly depend on the study’s objectives.

It is a major methodological challenge to detect genetic effects on brain structure and function, which are expected to have small effect sizes (Holland et al., Reference Holland, Wang, Thompson, Schork, Chen, Lo, Witoelar, Werge, O’Donovan, Andreassen and Dale2016; Mascarell Maričić et al., Reference Mascarell Maričić, Walter, Rosenthal, Ripke, Quinlan, Banaschewski, Barker, Bokde, Bromberg, Büchel, Desrivières, Flor, Frouin, Garavan, Itterman, Martinot, Martinot, Nees, Orfanos and Tahmasebi2020). Detecting these small effect sizes is further complicated by the small sample sizes and the so-called “curse of dimensionality” inherent to genetic and imaging data. The curse of dimensionality refers to the well-known problem where analytical methods fail to generalize as the parameter space exceeds the sample size. As the number of variables describing a sample increases relative to its participants, so does the chance that a model overfits the specific data and generates results that are less reliable and reproducible. Typically, imaging genetic datasets have very large feature spaces (e.g., hundreds of thousands) compared to the number of participants (e.g., a few hundred), making them susceptible to the curse of dimensionality unless addressed by the analytical strategy applied to the sample. Given the difficulties in collecting multimodal data from developing populations, it is difficult to generate large datasets for high-powered imaging genetic analysis. Thus, there is a need for large sample sizes to detect any associations between genes, brains, and development.

Large single-site cohorts can be difficult to achieve given the heterogeneity of neurodevelopmental disorders, variations in the manifestation and severity of such disorders, and the cost associated with the acquisition of both genetic and imaging data. Different methods have been proposed within each individual discipline to address some of these issues. A notable strategy in imaging contexts is the repeated acquisition of neuroimaging data for increasing reliability and significantly reducing the sample size needed to obtain a certain effect size (C. C. Fan et al., Reference Fan, Smeland, Schork, Chen, Holland, Lo, Sundar, Frei, Jernigan, Andreassen and Dale2018; McFarquhar et al., Reference McFarquhar, McKie, Emsley, Suckling, Elliott and Williams2016; Rosenberg & Finn, Reference Rosenberg and Finn2022). The issue of replicability is no surprise for anyone dealing with neurodevelopmental, neuroimaging, and genetic studies. Its complexity has been omnipresent. Thus, several large-scale studies were performed to provide evidence for and scope of the replicability of findings. The Open Science Collaboration (Open Science Collaboration, 2015) succeeded in replicating only 36 out of 100 prominent findings, whereas the Many Labs project not only replicated 10 of 13 across 36 independent samples but also demonstrated that replicability highly depends on the effect being studied (Klein et al., Reference Klein, Ratliff, Vianello, Adams, Bahník, Bernstein, Bocian, Brandt, Brooks, Brumbaugh, Cemalcilar, Chandler, Cheong, Davis, Devos, Eisner, Frankowska, Furrow, Galliani and Nosek2014), not so much on the specific sample or settings. These inconsistencies may be attributed to differences in sample characteristics, neuroimaging protocols, or statistical methods, highlighting the need for standardization and open science practice (Poldrack et al., Reference Poldrack, Baker, Durnez, Gorgolewski, Matthews, Munafò, Nichols, Poline, Vul and Yarkoni2017). For neurodevelopmental samples, the issue further increases from lack of homogeneity in the diagnosis of specific disorders, the difference in severity scales used, and the inclusion or exclusion of specific comorbidities, among others. These issues are then multiplied by the issues in neuroimaging data, which suffer from varying acquisition parameters across different sites (specifically for multisite larger datasets), quality of the data, and varying data analytic pipelines (Carter et al., Reference Carter, Bearden, Bullmore, Geschwind, Glahn, Gur, Meyer-Lindenberg and Weinberger2017).

Similar issues exist in genetic data analyses where biases can stem from various sources, potentially compromising the accuracy of genetic analysis. Allelic dropout (Johnson & Haydon, Reference Johnson and Haydon2007) is one common issue where one allele fails to amplify or is undetected during genotyping, leading to skewed results. Polymerase chain reaction (PCR) bias (Kebschull & Zador, Reference Kebschull and Zador2015) may occur due to factors like primer specificity or template secondary structure, causing certain alleles to be preferentially amplified over others. Reference bias (Günther et al., Reference Günther, Nettelblad and Di Rienzo2019) is another concern, as genotyping techniques rely on reference genomes or panels that may not fully represent the genetic diversity of the population being studied. Microarray-based genotyping can introduce chip bias (Kockum et al., Reference Kockum, Huang and Stridh2023), where factors like probe design or hybridization efficiency lead to inaccurate genotype calls. In NGS coverage, bias (Rieber et al., Reference Rieber, Zapatka, Lasitschka, Jones, Northcott, Hutter, Jäger, Kool, Taylor, Lichter, Pfister, Wolf, Brors, Eils and Hofmann2013) can arise, with some genomic regions being overrepresented or underrepresented due to uneven coverage. Genomic context bias is another issue, particularly in regions with repetitive sequences or high GC content (Lan et al., Reference Lan, Yin, Reed, Moua, Thomas and Zhang2015), which may be prone to amplification or sequencing errors. Sample preparation variability, including differences in DNA extraction or handling procedures, can also introduce biases in genotyping results. Following best practices for library preparation and genotyping, and being aware of these biases, is essential for accurately interpreting genetic data.

In response to the need for larger samples, there has been a growing number of databases that include brain, behavior, and genetic information for children. Large, open-source databases are becoming more common (see Boxes 3 and 4 for examples). These large datasets allow for more diverse populations to be studied and, importantly, allow for a more comprehensive view of child development by characterizing many child behaviors and, often, including many types of imaging data such as fMRI, sMRI, and diffusion MRI. Many of these databases have goals about diversity and include samples diverse in many characteristics, including race, SES, language background, and diagnosed developmental disorder, among other identifiers. While this diversity can make genetics studies more difficult, they are also arguably more representative of the current world than the typical W.E.I.R.D. (Western, educated, industrialized, rich, democratic) populations, which still rule many lab-based studies.

While the increase in data is a boon, researchers must be conscientious when performing analysis of heterogeneous populations. For example, there are natural differences in allele frequencies among populations with different genetic ancestry that must be accounted for in the analysis. Other confounders include environmental factors, such as exposure to toxins or stress; demographic variables, like age, sex, and ethnicity; and socioeconomic and sociocultural factors, like availability of resources, language exposure, and early childhood adversity. Failure to adequately account for these factors can lead to spurious statistical associations (Elliott et al., Reference Elliott, Sharp, Alfaro-Almagro, Shi, Miller, Douaud, Marchini and Smith2018; Lenroot & Giedd, Reference Lenroot and Giedd2008; Meyer-Lindenberg et al., Reference Meyer-Lindenberg, Nicodemus, Egan, Callicott, Mattay and Weinberger2008; Quinlan et al., Reference Quinlan, Barker, Luo, Banaschewski, Bokde, Bromberg, Büchel, Desrivières, Flor, Frouin, Garavan, Chaarani, Gowland, Heinz, Brühl, Martinot, Martinot, Nees, Orfanos and Schumann2020). Advanced statistical models, including mixed-effects models and path analysis, have been developed to address these confounding factors, making it possible to isolate the effect of genetic variants on neuroimaging outcomes (Chyzhyk et al., Reference Chyzhyk, Varoquaux, Milham and Thirion2022; Pala et al., Reference Pala, Lee, Ning, Kim and Shen2022). However, for various neurodevelopmental disorders with comorbidities, it is practically impossible to account for all confounding factors in a specific sample obtained for the study, which raises questions about the reliability of the findings (Mohammadi, Reference Mohammadi2015; Thompson et al., Reference Thompson, Stein, Medland, Hibar, Vasquez, Renteria, Toro, Jahanshad, Schumann and Franke2014). Therefore, it may be beneficial to focus on transdiagnostic symptom domains rather than focusing on specific disorders when using these large-scale resources, especially considering recent evidence indicating shared neural and genetic correlates across different disorders (e.g., Barker et al., Reference Barker, Ing, Biondo, Jia, Pingault, Du Rietz, Zhang, Ruggeri, Banaschewski, Hohmann, Bokde, Bromberg, Büchel, Quinlan, Sounga-Barke, Bowling, Desrivières, Flor, Frouin and Schumann2021; Hughes et al., Reference Hughes, Kunitoki, Elyounssi, Luo, Bazer, Hopkinson, Dowling, Doyle, Dunn, Eryilmaz, Gilman, Holt, Valera, Smoller, Cecil, Tiemeier, Lee and Roffman2023) Nonetheless, uniform guidelines that fit into an open science framework should be adopted to adequately mitigate the reproducibility crisis in imaging genetics research.

Statistical complexity and high-dimensional data

The complexity of genetic and brain imaging data that necessitates a large sample size also often requires advanced analytical techniques to parse out meaningful relationships, a task compounded by the vast heterogeneity observed within neurodevelopmental disorders (Arnett & Flaherty, Reference Arnett and Flaherty2022; Jacob et al., Reference Jacob, Wolff, Steinbach, Doyle, Kumar and Elison2019). In addition, the concern for privacy and the potential for genetic discrimination raises the ethical dilemma of this approach (Mezinska et al., Reference Mezinska, Gallagher, Verbrugge and Bunnik2021; Palk et al., Reference Palk, Illes, Thompson and Stein2020). Since many imaging genetic studies are interested in neurodevelopmental outcomes, extra care must be given to protecting the vulnerable populations under study. This includes equitable participant selection, appropriately informed consent/assent that reflects the growing necessity for these data to be stored in publicly accessible biobanks, and ethically reporting clinical results to participants and the broader public in a way that avoids psychological distress

(Palk et al., Reference Palk, Illes, Thompson and Stein2020). As global collaborations grow and generate large and diverse samples, researchers must be careful not to commit “ethics dumping,” the tendency to exploit potentially less regulated ethics oversight in low-to-middle-income countries, to expedite research aims (Palk et al., Reference Palk, Illes, Thompson and Stein2020).

The statistical complexity of analyzing these high-dimensional data is not handled well by traditional statistical methods. Methods that are used to analyze behavioral phenotypes, brain endophenotypes, and genetic variants concurrently to identify relationships between them must navigate these complex data while remaining robust to challenges posed by large sample sizes and high dimensionality (J. Fan et al., Reference Fan, Han and Liu2014). The development of appropriate methodology is dependent on available imaging genetic datasets suitable for testing such approaches. The first dataset suitable for testing such methods came from the Alzheimer’s Disease Neuroimaging Initiative, ADNI, and contained data modalities useful to imaging geneticists: clinical, behavioral, and cognitive assessments; brain imaging modalities, such as MRI and EEG; and genetic information such as genome-wide genotyping and whole-genome/exome sequencing (Weiner et al., Reference Weiner, Veitch, Aisen, Beckett, Cairns, Green, Harvey, Jack, Jagust, Morris, Petersen, Salazar, Saykin, Shaw, Toga and Trojanowski2017). The first large-scale whole-brain, whole-genome imaging genetics study was performed using these data in 2010, where researchers applied a massive univariate linear regression approach to identifying significant genetic markers associated with quantitative imaging indicators of AD progression (Stein et al., Reference Stein, Hua, Lee, Ho, Leow, Toga, Saykin, Shen, Foroud, Pankratz, Huentelman, Craig, Gerber, Allen, Corneveaux, DeChairo, Potkin, Weiner and Thompson2010).

Since then, numerous studies have explored analytical frameworks, ranging from multivariate regressions to machine learning to deep learning, suitable for jointly analyzing these combined data types. Although techniques from each discipline have proven effective in identifying significant associations between genes and imaging indicators, the prevailing approach is currently regularized multivariate regression due to its capacity to manage high-dimensional data within the constraints of current sample sizes, which typically include hundreds or thousands (Cheek et al., Reference Cheek, Lindner and Grigorenko2024). These regularized models have been the primary choice in large part because they are best able to mitigate the curse of dimensionality and prevent overfitting, as demonstrated in recent imaging genetic studies identifying genetic components of voxels associated with AD (Zhou et al., Reference Zhou, Thung, Liu and Shen2019). A major limiting factor for machine learning and deep learning techniques may be the number of available participants to train these models relative to the number of genetic and imaging features. However, as large studies and databanks begin to generate large cohorts of imaging genetic data (thousands of participants instead of hundreds), this paradigm may shift.

Network neuroscience

As new methods are developed, so too are new analytic frameworks. Network neuroscience represents a transformative approach to studying the neural mechanisms, shifting the focus from individual brain regions to the complex networks that underpin brain structural architecture, function, and cognition (Bassett & Sporns, Reference Bassett and Sporns2017; Koirala et al., Reference Koirala, Fleischer, Glaser, Zeuner, Deuschl, Volkmann, Muthuraman and Groppa2018; Michels et al., Reference Michels, Koirala, Groppa, Luechinger, Gantenbein, Sandor, Kollias, Riederer and Muthuraman2021). This new paradigm has significantly impacted imaging genetics, offering a different perspective to understanding the influence genetic variation has on the brain’s structural and functional networks, facilitating a deeper understanding of the polygenic nature of brain architecture. Unlike older imaging genetics studies focused on the association between individual genetic variants and specific brain regions or functions, network neuroscience has highlighted the importance of considering how the genome influences the brain’s overall connectivity. For example, studies using graph theory metrics such as node degree, clustering coefficient, and path length have shown that genetic factors can impact the efficiency and resilience of neural networks (Dennis et al., Reference Dennis, Jahanshad, Rudie, Brown, Johnson, McMahon, de Zubicaray, Montgomery, Martin, Wright, Bookheimer, Dapretto, Toga and Thompson2011; Fornito & Bullmore, Reference Fornito and Bullmore2012). These metrics provide a quantitative framework to assess the genetic contribution to the brain’s organizational principles and offer insights into the genetic bases of individual differences in cognitive abilities, emotional processing, and susceptibility to mental health disorders. Similarly, network neuroscience has enabled the identification of genetic influences on the brain’s modular organization. The human brain is organized into functionally specialized but interdependent networks, such as the default mode network, salience network, and central executive network (Power et al., Reference Power, Cohen, Nelson, Wig, Barnes, Church, Vogel, Laumann, Miezin, Schlaggar and Petersen2011; Yeo et al., Reference Yeo, Krienen, Sepulcre, Sabuncu, Lashkari, Hollinshead, Roffman, Smoller, Zöllei, Polimeni, Fischl, Liu and Buckner2011). Imaging genetics studies incorporating network analysis have demonstrated that specific genetic variants can modulate the coherence and integration of these networks. For instance, variations in genes related to neurotransmitter systems, such as the dopamine and serotonin receptors, have been associated with alterations in the connectivity within and between these functional networks (Gaiteri et al., Reference Gaiteri, Ding, French, Tseng and Sibille2014; Parikshak et al., Reference Parikshak, Gandal and Geschwind2015; Vértes et al., Reference Vértes, Rittman, Whitaker, Romero-Garcia, Váša, Kitzbichler, Wagstyl, Fonagy, Dolan, Jones, Goodyer and Bullmore2016). This has profound implications for understanding the genetic underpinnings of complex behaviors and neuropsychiatric conditions, as disruptions in network connectivity and integration are a common feature of disorders such as schizophrenia, ASD, and MDD (Benkarim et al., Reference Benkarim, Paquola, Park, Hong, Royer, Vos de Wael, Lariviere, Valk, Bzdok, Mottron and C. Bernhardt2021; Duman et al., Reference Duman, Sanacora and Krystal2019; Gulsuner et al., Reference Gulsuner, Walsh, Watts, Lee, Thornton, Casadei, Rippey, Shahin, Nimgaonkar, Go, Savage, Swerdlow, Gur, Braff, King, McClellan, Braff, Cadenhead, Calkins and Wilson2013). Recently, network neuroscience approaches have been employed to investigate how genetic predispositions interact with environmental exposures to shape the development and plasticity of neural networks (Flores-Dorantes et al., Reference Flores-Dorantes, Díaz-López and Gutiérrez-Aguilar2020; J.-W. Kim et al., Reference Kim, Park, Kang, Gonzales, Oh, Seung, Ko, Cheong, Chung and Shin2019; Tooley et al., Reference Tooley, Bassett and Mackey2021), which is critical for understanding the complex interplay between genetics, environment, and brain function, informing interventions, and developing personalized therapeutic strategies for neurodevelopmental and psychiatric disorders.