Periodic Reporting for period 2 - CONNECT (Connecting cross-condition patterns of brain connectivity towards a common mechanism of mental conditions and prediction connectomics)
Reporting period: 2023-03-01 to 2024-08-31
Brain disorders are marked by unique features yet share many clinical and genetic associations. Numerous neuroimaging studies have examined the brain morphological and connectivity patterns associated with these conditions, however, there is a scarcity of studies aiming to identify disorder-specific and cross-disorder brain alterations through a unified approach across disorders. The ERC CONNECT project strives to map and understand brain mechanisms underlying the overlap and specificities between different neuropsychiatric diseases. To achieve this aim, this project is structured around five main objectives covering both logistic and research aspects. These objectives comprise the pooling, preprocessing, and processing of a large number of (open) datasets across a wide range of neurologic and psychiatric brain conditions, the development of methodologies and tools to allow us to curate, combine, analyze and compare large datasets, derive and compare disease fingerprints, and the development of new machine learning based algorithms and tools to derive meaningful metrics for potential future clinical use.
Understanding how the brain functions in healthy individuals and in patients with brain disorders is a continuously pressing medical challenge. Traditionally, researchers viewed brain and mental disorders as distinct entities. While this approach helps diagnose and treat patients, recent evidence suggests that many disorders may share common underlying mechanisms, with patterns of brain alterations shared across conditions.
Magnetic resonance imaging (MRI) has become a powerful tool for studying the structure and function of the human brain. The influence of network neuroscience –the study of the relationships in structural and functional connections between brain regions– on mental health has grown to be an important area of research and a large body of neuroimaging studies have identified a wide range of abnormalities in brain connectivity for various mental disorders. These MRI studies often study these abnormalities for different conditions, mostly in isolation of other conditions. Less is known about how such brain patterns may overlap and/or differentiate. The CONNECT project now aims to bring together these datasets and investigate relationships in disease-related brain alterations between and across mental disorders.
Our project aims to identify common biological mechanisms by building a large-scale MRI database encompassing multiple mental disorders. This data is used to develop a framework to systematically search for shared patterns of brain function and dysfunction. CONNECT will further differentiate and disentangle these commonalities from disorder-specific alterations. The project is advancing our understanding in the fields of neuroimaging and network neuroscience through network analyses of neuroimaging data across various neuropsychiatric and neurological disorders. So far, we have achieved the collection, processing, and analysis of large volumes of MRI datasets (120 datasets, including over 26,000 controls and 15,000 patients across 20 disorders, in addition to the large-scale UK Biobank comprising 37,000 individuals, encompassing 15,000 patients with neuropsychiatric disorders) through means of open datasets and the establishment of invaluable academic collaborations. The data is used for the identification of brain patterns associated with each of these specific disorders, bringing 20 different disease brainmaps. Our next step will be to compare the derived disease maps and examine dysconnectivity cross-disorders patterns. Our main research objectives include assessing the replicability of neuroimaging findings, investigating cross-condition connectomic features, developing novel statistical tools (the PolyConnectomic Risk Score) for network analysis, and the creation of a repository of connectomic disease signatures to be utilized by the scientific community. Through these collaborations and developments of various methodologies, the project contributes significantly to the field to facilitate a more enhanced picture of brain disorders.
Understanding how the brain functions in healthy individuals and in patients with brain disorders is a continuously pressing medical challenge. Traditionally, researchers viewed brain and mental disorders as distinct entities. While this approach helps diagnose and treat patients, recent evidence suggests that many disorders may share common underlying mechanisms, with patterns of brain alterations shared across conditions.
Magnetic resonance imaging (MRI) has become a powerful tool for studying the structure and function of the human brain. The influence of network neuroscience –the study of the relationships in structural and functional connections between brain regions– on mental health has grown to be an important area of research and a large body of neuroimaging studies have identified a wide range of abnormalities in brain connectivity for various mental disorders. These MRI studies often study these abnormalities for different conditions, mostly in isolation of other conditions. Less is known about how such brain patterns may overlap and/or differentiate. The CONNECT project now aims to bring together these datasets and investigate relationships in disease-related brain alterations between and across mental disorders.
Our project aims to identify common biological mechanisms by building a large-scale MRI database encompassing multiple mental disorders. This data is used to develop a framework to systematically search for shared patterns of brain function and dysfunction. CONNECT will further differentiate and disentangle these commonalities from disorder-specific alterations. The project is advancing our understanding in the fields of neuroimaging and network neuroscience through network analyses of neuroimaging data across various neuropsychiatric and neurological disorders. So far, we have achieved the collection, processing, and analysis of large volumes of MRI datasets (120 datasets, including over 26,000 controls and 15,000 patients across 20 disorders, in addition to the large-scale UK Biobank comprising 37,000 individuals, encompassing 15,000 patients with neuropsychiatric disorders) through means of open datasets and the establishment of invaluable academic collaborations. The data is used for the identification of brain patterns associated with each of these specific disorders, bringing 20 different disease brainmaps. Our next step will be to compare the derived disease maps and examine dysconnectivity cross-disorders patterns. Our main research objectives include assessing the replicability of neuroimaging findings, investigating cross-condition connectomic features, developing novel statistical tools (the PolyConnectomic Risk Score) for network analysis, and the creation of a repository of connectomic disease signatures to be utilized by the scientific community. Through these collaborations and developments of various methodologies, the project contributes significantly to the field to facilitate a more enhanced picture of brain disorders.
Over 120 multimodal MRI brain datasets have been collected and jointly analyzed. Those datasets encompass approximately 15,000 patients (and another 26,000 healthy controls to compare with) across 20 neuropsychiatric disorders. The collection of those datasets was possible not only due to large collaborative efforts for open sharing of data, but also through direct collaborations strengthened and enabled by this project. A complete list of the included datasets are presented in the Data Management Plan of CONNECT.
In the first part of the project, all data has been uniformly processed using a software package called Connectivity Analysis TOolbox (CATO), developed in our lab as part of the CONNECT project (de Lange et al. 2023). This data preprocessing package consists of a processing pipeline for resting-state functional MRI (rs-fMRI) and diffusion weighted imaging (DWI) data and has been made publicly available along with documentation on how to install and use it. The application of this pipeline across all of our datasets is one of the relevant steps that we take to achieve reliability and replicability of subsequent analysis results. The CATO package is made publicly available.
The next goal of the CONNECT project is to use all the analyzed individual datasets to build group-based disease maps, describing the altered morphological, structural and functional connectivity changes related to disease. Currently, of 6 major disorders first versions of disease maps have been created. These maps are part of a pre-preprint scientific paper and will be made public by means of a public toolbox (manuscript in preparation).
As part of this process, we have empirically investigated the relationship between sample size and replicability of neuroimaging findings in brain disorders, to obtain insight into the relative robustness of the build disease maps. For each of the six conditions, multiple datasets are used and compared to obtain metrics of reliability and robustness of the presented maps.
We have further performed a study on the state of the art of statistical analyses in the field, in particular with a focus on ‘statistical power’ (the ability of a study to find robust effects). We provided a list of guidelines to the community to improve reliability in the field, and a ‘how-to’ of obtaining insight in the quality of used datasets. This review covers the impact of statistical power on neuroimaging findings across different levels of analysis, where power can have an influence at the study level but also at the connection level. The review has been published (Helwegen et al. 2023).
Disease maps (i.e. comprehensive representations of altered functional connections within the brain associated with various psychiatric and neurological conditions) derived from aggregated data and standardized analyses, have been computed for each disorder across three modalities: brain morphology, functional connectivity, and structural connectivity. These maps reveal patterns of both hypoconnectivity (reduced in disease) and hyperconnectivity (increased in disease) across disorders. As such, they now facilitate the exploration of both the overlap and heterogeneity in connectivity changes related to disease.
A third achievement of the project so far is the development of the ‘polyconnectomic score (PCS)’, a metric designed to calculate a disorder-specific risk estimate for individuals based on their whole-brain connectomic profile. Inspired by the conceptual framework of a genetic polygenic scores in the field of genetics, the PCS aims to quantitatively encapsulate disease-related connectivity manifestations within individual connectomes. The formulation and validation of the PCS metric has been undertaken within the domain of functional connectivity analysis. Drawing upon the rich landscape of network neuroscience, PCS amalgamates findings from large-scale connectomic datasets to address the inherent challenge posed by low sample and effect size, as well as the proliferating number of statistical tests encumbered during conventional analyses. Our PCS paper is presented as a pre-print (doi: 10.1101/2023.09.26.559327) and is currently discussed at scientific meetings (DNM 2023, OHBM 2024).
In the first part of the project, all data has been uniformly processed using a software package called Connectivity Analysis TOolbox (CATO), developed in our lab as part of the CONNECT project (de Lange et al. 2023). This data preprocessing package consists of a processing pipeline for resting-state functional MRI (rs-fMRI) and diffusion weighted imaging (DWI) data and has been made publicly available along with documentation on how to install and use it. The application of this pipeline across all of our datasets is one of the relevant steps that we take to achieve reliability and replicability of subsequent analysis results. The CATO package is made publicly available.
The next goal of the CONNECT project is to use all the analyzed individual datasets to build group-based disease maps, describing the altered morphological, structural and functional connectivity changes related to disease. Currently, of 6 major disorders first versions of disease maps have been created. These maps are part of a pre-preprint scientific paper and will be made public by means of a public toolbox (manuscript in preparation).
As part of this process, we have empirically investigated the relationship between sample size and replicability of neuroimaging findings in brain disorders, to obtain insight into the relative robustness of the build disease maps. For each of the six conditions, multiple datasets are used and compared to obtain metrics of reliability and robustness of the presented maps.
We have further performed a study on the state of the art of statistical analyses in the field, in particular with a focus on ‘statistical power’ (the ability of a study to find robust effects). We provided a list of guidelines to the community to improve reliability in the field, and a ‘how-to’ of obtaining insight in the quality of used datasets. This review covers the impact of statistical power on neuroimaging findings across different levels of analysis, where power can have an influence at the study level but also at the connection level. The review has been published (Helwegen et al. 2023).
Disease maps (i.e. comprehensive representations of altered functional connections within the brain associated with various psychiatric and neurological conditions) derived from aggregated data and standardized analyses, have been computed for each disorder across three modalities: brain morphology, functional connectivity, and structural connectivity. These maps reveal patterns of both hypoconnectivity (reduced in disease) and hyperconnectivity (increased in disease) across disorders. As such, they now facilitate the exploration of both the overlap and heterogeneity in connectivity changes related to disease.
A third achievement of the project so far is the development of the ‘polyconnectomic score (PCS)’, a metric designed to calculate a disorder-specific risk estimate for individuals based on their whole-brain connectomic profile. Inspired by the conceptual framework of a genetic polygenic scores in the field of genetics, the PCS aims to quantitatively encapsulate disease-related connectivity manifestations within individual connectomes. The formulation and validation of the PCS metric has been undertaken within the domain of functional connectivity analysis. Drawing upon the rich landscape of network neuroscience, PCS amalgamates findings from large-scale connectomic datasets to address the inherent challenge posed by low sample and effect size, as well as the proliferating number of statistical tests encumbered during conventional analyses. Our PCS paper is presented as a pre-print (doi: 10.1101/2023.09.26.559327) and is currently discussed at scientific meetings (DNM 2023, OHBM 2024).
The CONNECT project presents a state of the art framework in the field of network neuroscience in terms of size of the datasets combined, number of disorders investigated, comprehensive coverage of complementary neuroimaging modalities (e.g. brain morphology, structural connectivity, and functional connectivity) and the sample sizes collected within each disorder, all with the joint goal of identifying common macroscale mechanisms of neuropsychiatric diseases. The next step (phase II of the project) is to curate and clean our collated repository of neuroimaging data that will allow for us to move forward in the identification of common and specific disease alteration patterns associated with particular brain disorders. This next step of the project is important for understanding neuropathological processes underlying a wide range of brain disorders, as well as the potential to develop brain markers specific for neuropsychiatric conditions.
Additionally, the project will also further contribute to the methodology in the field as a whole; a primary focus in network neuroscience is addressing critical issues such as replicability of neuroimaging findings and adequacy of statistical power, both of which are fundamental concerns in this domain. By standardizing data collection protocols and implementing robust statistical frameworks, our project is making notable strides towards enhancing reliability and reproducibility of neuroimaging analyses. A prime example of this advancement is the proposed introduction of the machine-learning based polyconnetomic score that examines the use of more personalized assessments of an individual's brain connectivity by offering a perspective on an individual's neural architecture. This metric not only provides a disorder-specific risk estimate based on comprehensive connectomic profiles but also lays the groundwork for correlating brain signatures to clinical and behavioral measurements thus enabling a deeper understanding of neurobiological underpinnings within and across various neuropsychiatric conditions. We will continue the development of the PCS in the second phase of the project and in particular test the level of specificity and predictive power. We will test PCS across multiple datasets, as well as within and between conditions. Additional projects involve linking PCS and brain metrics to cross-disorder genetics, and using machine learning techniques to examine the translation of cross-disorder disease-common and disease-specific patterns for clinical diagnosis. We are also in the process of incorporating additional MRI atlas parcellation schemes into our pipelines to increase the coverage and standardization of human brain mapping, particularly in regions that have historically been neglected, such as the subcortex and cerebellum. In examining differences in parcellation schemes, we will assess how choice of atlas impacts downstream analyses and findings, allowing us to evaluate robustness and consistency of results across different parcellation methods. During this process, factors such as biological plausibility of parcellations, reproducibility of findings, and generalizability to other populations or pathologies will be carefully considered. We also aim to further examine gender-specific disease patterns, examining the effects of patient-subpopulation on within- and cross-disease patterns.
Additionally, the project will also further contribute to the methodology in the field as a whole; a primary focus in network neuroscience is addressing critical issues such as replicability of neuroimaging findings and adequacy of statistical power, both of which are fundamental concerns in this domain. By standardizing data collection protocols and implementing robust statistical frameworks, our project is making notable strides towards enhancing reliability and reproducibility of neuroimaging analyses. A prime example of this advancement is the proposed introduction of the machine-learning based polyconnetomic score that examines the use of more personalized assessments of an individual's brain connectivity by offering a perspective on an individual's neural architecture. This metric not only provides a disorder-specific risk estimate based on comprehensive connectomic profiles but also lays the groundwork for correlating brain signatures to clinical and behavioral measurements thus enabling a deeper understanding of neurobiological underpinnings within and across various neuropsychiatric conditions. We will continue the development of the PCS in the second phase of the project and in particular test the level of specificity and predictive power. We will test PCS across multiple datasets, as well as within and between conditions. Additional projects involve linking PCS and brain metrics to cross-disorder genetics, and using machine learning techniques to examine the translation of cross-disorder disease-common and disease-specific patterns for clinical diagnosis. We are also in the process of incorporating additional MRI atlas parcellation schemes into our pipelines to increase the coverage and standardization of human brain mapping, particularly in regions that have historically been neglected, such as the subcortex and cerebellum. In examining differences in parcellation schemes, we will assess how choice of atlas impacts downstream analyses and findings, allowing us to evaluate robustness and consistency of results across different parcellation methods. During this process, factors such as biological plausibility of parcellations, reproducibility of findings, and generalizability to other populations or pathologies will be carefully considered. We also aim to further examine gender-specific disease patterns, examining the effects of patient-subpopulation on within- and cross-disease patterns.