Christian Friedrich Beckmann

Digital phenotyping, which is defined as quantifying someone’s behaviour with digital devices, provides unprecedented opportunities for understanding human mental health but is hampered by high levels of inter-individual variability. Here, we propose a new method to address this, parsing inter-individual variability by decomposing the digital phenotype dynamics into latent trajectories and using each individual’s trajectory membership as a moderator when modelling psychopathology over the same timeframe. We applied our method in the context of mood symptom exacerbation across the menstrual cycle, where symptom severity and timing are inconsistent between individuals. Using the BiAffect platform to collect smartphone typing dynamics, we found stable trajectories in smartphone movement rate: one group of participants showed substantial movement rate fluctuations across the menstrual cycle, whilst the others did not. Participants with movement fluctuations displayed increased fluctuations across the cycle in prospective anhedonia and depression ratings, but not in anxiety, irritability, and suicidal ideation.

Realigning structural MR images to stereotactic AC-PC space is a standard procedure that enhances anatomical consistency both within and across neuroimaging studies, allowing for precise spatial localization and reliable cross-subject comparisons in both clinical and research contexts. However, different versions of AC-PC spaces, stemming from varying definitions of the AC-PC axis as outlined in common stereotactic atlases like Talairach and Schaltenbrand, often lead to discrepancies between the neurosurgical and neuroimaging communities. Manual realignment of structural MR images to the clinically used version of the AC-PC plane is often necessary to validate the results of automated methods or to realign the image when these methods fail. Furthermore, manual realignment provides a critical ‘ground truth’ for the development of new automated tools. However, such manual interventions are typically performed in a non-standardised manner by domain experts who possess the specialized knowledge of neuroanatomy and neuroimaging required to ensure accurate alignment. To address these challenges, we have developed and validated a standardized protocol for manually realigning structural MR images to the clinically used Schaltenbrand AC-PC plane, using a set of visual criteria to ensure accurate realignment. This protocol can be used to manually align images, verify results, validate existing automated methods, or generate ground truth data for developing new automated techniques.

The uniquely human capacity for language is supported by a distributed brain network with cerebral cortical hubs in the left inferior frontal gyrus (LIFG) and left superior temporal sulcus (LSTS). Individual differences in the graded spatial organization of function across these regions may provide sensitive phenotypes for behavioural and genetic association analyses, to better understand the biological underpinnings of language. In 41,437 UK Biobank participants we applied connectopic mapping to the LIFG and LSTS. This approach maps spatial changes across a region in terms of connectivity with the rest of the brain, based on resting state functional imaging. On average, we found that Brodmann Area (BA) 44, BA 45 and the posterior LSTS were strongly connected to the superior and inferior parietal lobe and the dorsal attention and frontoparietal networks, whereas BA 47 and the medial LSTS were strongly connected to superior frontal areas BA 8B and BA 9, and the default mode network more broadly. However, there was extensive population-wide variation in these graded connectopic patterns, which significantly mediated the effects of polygenic scores for reading ability and educational attainment on performance on a vocabulary task. Furthermore we identified 3 genomic loci displaying multivariate association patterns with these connectopic maps, one of which coincides with a long non-coding RNA, LINC01165 , with potential relevance in human evolution. Overall, connectopic mapping produced fewer significant genetic association findings and lower heritability (range from 0.3 to 9.4%) than previous parcel-based approaches for mapping language network connectivity. These observations suggest higher genetic complexity and/or prominent roles for environmental or random developmental effects on these connectopic maps.

Population-scale neuroimaging allows for novel biological discovery, but voxelwise analyses are computationally paralyzing and noisy, whereas imaging-derived phenotypes discard crucial spatial detail. We introduce PANDORA (Population Archive of Neuroimaging Data Organized for Rapid Analysis), a data-adaptive modelling platform designed to resolve this trade-off. PANDORA has encoded brain MRI data comprising 98 sub-modalities from over 80,000 UK Biobank participants in a highly efficient supervoxel representation. By performing statistical regressions directly within this compressed embedding, PANDORA reduces storage by up to 99% and accelerates computation 10-fold, while acting as a spatial denoiser to enhance statistical power. PANDORA also includes the full-resolution voxelwise ground-truth data, curated imaging confound variables, and a fast analysis tool achieving whole brain, voxelwise population-level regression in seconds to minutes. We showcase PANDORA’s ability to reproduce known patterns and reveal new associations including trauma, anxiety/depression, autism polygenic scores, and EPHA3 .

[This corrects the article DOI: 10.2196/64007.].

Brain structure and function have largely been studied separately in relation to the neurobiology and genetics of language. Here we used linked independent component analysis to integrate language network functional connectivity with brain volumetric and white matter structure in 32,677 UK Biobank participants, followed by analysis of behavioural, neurobiological and genetic correlates of the derived multimodal structure-function imaging components. Stronger functional connectivity between brain language areas was associated with increased volume of parts of the cerebellum and motor cortex, together with smaller ventricles and sensory parietal and occipital areas. The brain structure-function language components mediated an association between vocabulary level and polygenic scores for reading ability. We report 18 genomic loci associated with brain structure-function language components. Single-nucleotide polymorphism (SNP)-based heritability estimates for these components were 23-30%, and there was significant enrichment of heritability in primate-conserved genomic loci and fetal brain human-gained enhancer elements. This study revealed that structural correlates of functional language network connectivity extend well beyond previously defined language areas of the brain, and highlights the value of multimodal brain phenotyping for human neurogenetic discovery.

INTRODUCTION: Alzheimer’s disease (AD) is heterogeneous, complicating diagnosis and prognosis. Uncovering patterns that link abnormalities across amyloid-tau-neurodegeneration (A-T-N) framework may improve prediction of clinical diagnosis. METHODS: We applied SuperBigFLICA (SBF), a semi-supervised multimodal data fusion method, to maps of gray matter density, cortical thickness, pial surface area, amyloid PET, and tau PET in 274 ADNI-3 participants. The model was trained to derive 50 latent components most predictive of a continuous measure of cognitive decline. Latent components’ subject loadings were subsequently used to predict diagnosis (cognitively normal, mild cognitive impairment, dementia) and APOE4 status using LASSO logistic regression, and were compared against demographic, single-modality, and naïve fusion baselines. RESULTS: While SuperBigFLICA modestly predicted cognitive decline (r = 0.21), SBF loadings-based models outperformed baselines (AUROC = 0.80 for diagnosis; 0.83 for APOE4 ). Amyloid alterations in sensory areas along the sensory–association axis best separated dementia, while a multimodal A-T-N pattern was related to early cognitive decline. Subject loadings on these two patterns were associated with cerebrospinal fluid (CSF) biomarkers, highlighting how CSF AD biomarkers relate to spatial patterns of brain A-T-N burden. DISCUSSION: Semi-supervised multimodal fusion improves prediction and reveals interpretable imaging patterns that predict APOE4 and clinical diagnoses better than traditional approaches.

Traditional group-level fMRI analysis approaches, such as Independent Component Analysis (ICA), often rely on unsupervised dimensionality reduction to map subjects into a common feature space. While effective for capturing common variance across all subjects, the preservation of discriminative features between groups of participants is not guaranteed. To address this limitation, we introduce Independent Filter Analysis (IFA), a supervised extension of group ICA that explicitly models group-discriminative information as part of the dimensionality reduction steps. Prior to unmixing, IFA constructs a subspace that simultaneously retains both shared and group-specific information, enhancing sensitivity to group effects while preserving biological interpretability. We validated IFA using simulated data and paired condition comparisons from three Human Connectome Project (HCP) tasks. In the simulation, IFA achieved 95% classification accuracy, outperforming group ICA, which failed to detect subtle group differences. On the HCP data, IFA increased network matrix classification accuracy by up to 15% and produced spatial maps that more precisely reflected task-relevant differences.

UNSTRUCTURED Please see correction word document.

In functional magnetic resonance imaging, multivariate proxies of functional brain networks are commonly extracted using spatial independent component analysis. The theoretical premises of spatial overlap among functional processes and the time-varying nature of functional connectivity prompt the question of how to accurately model spatially overlapping and time-varying functional sources. Well-known functional networks have previously been shown to divide into spatially overlapping and functionally distinct subprocesses termed Temporal Functional Modes (TFM) using temporal independent component analysis on the time courses obtained via spatial independent component analysis. In this model, spatial and temporal modes of organisation interact through a single mixing matrix with fixed coefficients. Here, we introduce a time-resolved version termed Time-Resolved Instantaneous Functional Loci Estimation (TRIFLE) to estimate time-varying changes in source allocation. We analytically demonstrate that the originally fixed TFM mixing matrix can be expressed as the temporal average of a time-resolved mixing matrix, which in turn can be obtained in closed form and provides instantaneous estimates of brain network reconfigurations involved in distinct temporal functional modes. We apply TRIFLE to a high-temporal resolution functional magnetic resonance imaging dataset. We demonstrate that spatial source allocation aligns with expectations based on the experimental task design and that successful and unsuccessful trials have different allocation profiles. The proposed method sheds light on the temporal evolution of brain network reconfigurations while explicitly accounting for potential neuroanatomical overlap.