Biology subjects

Mansour L., S.

Publications and source records attributed to Mansour L., S..

7 recordsLinked to original sources

bioRxiv2026

Widespread use of invalid statistical tests in biomedical machine learning

Cross-validation is routinely used to compare performance in biomedical artificial intelligence. Standard tests ignore correlation across cross-validation folds, inflating false-positive rates. In a PRISMA-guided meta-analysis of 184 studies (impact factor [≥] 15) across 30 biomedical fields, 97% use invalid tests. Among studies with abstract-level claims supported by invalid tests, 59% rely on a spurious comparison: one that loses significance after correcting for this correlation. On average, regaining significance requires a 43% larger effect. Extending these findings to the broader literature, we estimate that spurious comparisons occur in nearly two in three studies and support abstract-level claims in nearly one in three studies. Simulations confirm false-positive rates paradoxically approach 100% when cross-validation is repeated to improve stability. We introduce SHARP, a redesign of cross-validation, which best balances false-positive control and power among 13 benchmarked tests. These results reveal widespread fragility in biomedical artificial intelligence and offer a practical route to valid comparisons.

bioinformatics↗

bioRxiv2025

Deep generation of personalized connectomes based on individual attributes

An individuals connectome is unique. Interindividual variation in connectome architecture associates with disease status, cognition, lifestyle factors, and other personal attributes. While models to predict personal attributes from a persons connectome are abundant, the inverse task-- inferring connectome architecture from an individuals personal profile--has not been widely studied. Here, we introduce a deep model to generate a persons entire connectome exclusively based on their age, sex, body phenotypes, cognition, and lifestyle factors. Using the richly phenotyped UK Biobank connectome cohort (N=8,086), we demonstrate that our model can generate network architectures that closely recapitulate connectomes mapped empirically using diffusion MRI and tractography. We find that age, sex, and body phenotypes exert the strongest influence on the connectome generation process, with an impact approximately four times greater than that of cognition and lifestyle factors. Regional differences in the importance of measures were observed, including an increased importance of cognition in the association cortex relative to the visual system. We further show that generated connectomes can improve the training of machine learning models and reduce their predictive errors. Our work demonstrates the feasibility of inferring brain connectivity from an individuals personal data and enables future applications of connectome generation such as data augmentation and anonymous data sharing.

neuroscience↗

bioRxiv2024

On reconstruction of cortical functional maps using subject-specific geometric and connectome eigenmodes

Understanding the interplay between human brain structure and function is crucial to discern neural dynamics. This study explores the relation between brain structure and macroscale functional activity using subject-specific structural connectome eigenmodes, complementing prior work that focused on group-level models and geometry. Leveraging data from the Human Connectome Project, we assess accuracy in reconstructing various functional MRI-based cortical maps using individualised eigenmodes, specifically, across a range of connectome construction parameters. Our results show only minor differences in performance between surface geometric eigenmodes, a local neighborhood graph, a highly smoothed null model, and individual and group-level connectomes at modest smoothing and density levels. Furthermore, our results suggest that spatially smooth eigenmodes best explain functional data. The absence of improvement of individual connectomes and surface geometry over smoothed null models calls for further methodological innovation to better quantify and understand the degree to which brain structure constrains brain function.

neuroscience↗

bioRxiv2024

Development of thalamocortical connectivity during the third trimester

Thalamocortical connections are crucial for relaying sensory information in the brain and facilitate essential functions including motor skills, emotion, and cognition. Emerging evidence suggests that thalamocortical connections are organised along spatial gradients that may reflect their sequential formation during early brain development. However, this has not been extensively characterised in humans. To examine early thalamocortical development, we analysed diffusion MRI data from 345 infants, scanned between 29-45 weeks gestational age. Using diffusion tractography, we mapped thalamocortical connectivity in each neonate and used Principal Component Analysis to extract shared spatial patterns of connectivity. We identified a primary axis of connectivity that varied along an anterior/medial to posterior/lateral gradient within the thalamus, with corresponding projections to cortical areas varying along a rostral-caudal direction. The primary patterns of thalamocortical connectivity were present at 30 weeks gestational age and gradually refined during gestation. This refinement was largely driven by the maturation of connections between the thalamus and cortical association areas. Differences in thalamocortical connectivity between preterm and term neonates were only weakly related to primary thalamocortical gradients, suggesting a relative preservation of these features following premature birth. Overall, our results indicate that the organisation of structural thalamocortical connections are highly conserved across individuals, develop early in gestation and gradually mature with age.

neuroscience↗

bioRxiv2024

Eigenmodes of the brain: revisiting connectomics and geometry

Eigenmodes can be derived from various structural brain properties, including cortical surface geometry1 and interareal axonal connections comprising an organisms connectome2. Pang and colleagues map geometric and connectome eigenmodes to spatial patterns of human brain activity, assessing whether brain connectivity or geometry provide greater explanatory power of brain function3. The authors find that geometric eigenmodes are superior predictors of cortical activity compared to connectome eigenmodes. They conclude that this supports the predictions of neural field theory (NFT)4, in that "brain activity is best represented in terms of eigenmodes derived directly from the shape of the cortex, thus emphasizing a fundamental role of geometry in constraining dynamics". The experimental comparisons favoring geometric eigenmodes over connectome eigenmodes, in conjunction with specific statements regarding the relative efficacy of geometry in representing brain activity, have been widely interpreted to mean that geometry imposes stronger constraints on cortical dynamics than connectivity5-9. Here, we reconsider the comparative experimental evidence focusing on the impact of connectome mapping methodology. Utilizing established methods to mitigate connectome construction limitations, we map new connectomes for the same dataset, finding that eigenmodes derived from these connectomes reach comparable accuracy in explaining brain activity to that of geometric eigenmodes. We conclude that the evidence presented to support the comparative proposition that "eigenmodes derived from brain geometry represent a more fundamental anatomical constraint on dynamics than the connectome" may require reconsideration in light of our findings. Pang and colleagues present compelling evidence for the important role of geometric constraints on brain function, but their findings should not be interpreted to mean that geometry has superior explanatory power over the connectome.

neuroscience↗

bioRxiv2023

Connectomes for 40,000 UK Biobank participants: A multi-modal, multi-scale brain network resource

We mapped functional and structural brain networks for more than 40,000 UK Biobank participants. Structural connectivity was estimated with tractography and diffusion MRI. Resting-state functional MRI was used to infer regional functional connectivity. We provide high-quality structural and functional connectomes for multiple parcellation granularities, several alternative measures of interregional connectivity, and a variety of common data pre-processing techniques, yielding more than one million connectomes in total and requiring more than 200,000 hours of compute time. For a single subject, we provide 28 out-of-the-box versions of structural and functional brain networks, allowing users to select, e.g., the parcellation and connectivity measure that best suit their research goals. Furthermore, we provide code and intermediate data for the time-efficient reconstruction of more than 1,000 different versions of a subjects connectome based on an array of methodological choices. All connectomes are available via the UK Biobank data sharing platform and our connectome mapping pipelines are openly available. In this report, we describe our connectome resource in detail for users, outline key considerations in developing an efficient pipeline to map an unprecedented number of connectomes, and report on the quality control procedures that were completed to ensure connectome reliability and accuracy. We demonstrate that our structural and functional connectivity matrices meet a number of quality control checks and replicate previously established findings in network neuroscience. We envisage that our resource will enable new studies of the human connectome in health, disease and aging at an unprecedented scale.

neuroscience↗

bioRxiv2021

Connectome-Based Smoothing (CBS): Concepts, methods, and evaluation

Structural connectomes are increasingly mapped at high spatial resolutions comprising many hundreds--if not thousands--of network nodes. However, high-resolution connectomes are particularly susceptible to image registration misalignment, tractography artifacts, and noise, all of which can lead to reductions in connectome accuracy and test-retest reliability. We investigate a network analogue of image smoothing to address these key challenges. Connectome Spatial Smoothing (CSS) involves jointly applying a carefully chosen smoothing kernel to the two endpoints of each tractography streamline, yielding a spatially smoothed connectivity matrix. We develop computationally efficient methods to perform CSS using a matrix congruence transformation and evaluate a range of different smoothing kernel choices on CSS performance. We find that smoothing substantially improves the identifiability, sensitivity, and test-retest reliability of high-resolution connectivity maps, though at a cost of increasing storage burden. For atlas-based connectomes (i.e. low-resolution connectivity maps), we show that CSS marginally improves the statistical power to detect associations between connectivity and cognitive performance, particularly for connectomes mapped using probabilistic tractography. CSS was also found to enable more reliable statistical inference compared to connectomes without any smoothing. We provide recommendations on optimal smoothing kernel parameters for connectomes mapped using both deterministic and probabilistic tractography. We conclude that spatial smoothing is particularly important for the reliability of high-resolution connectomes, but can also provide benefits at lower parcellation resolutions. We hope that our work enables computationally efficient integration of spatial smoothing into established structural connectome mapping pipelines. HighlightsO_LIWe establish a network equivalent of image smoothing for structural connectomes. C_LIO_LIConnectome Spatial Smoothing (CSS) improves connectome test-retest reliability, identifiability and sensitivity. C_LIO_LICSS also facilitates reliable inference and improves power to detect statistical associations. C_LIO_LIBoth high-resolution and atlas-based connectomes can benefit from CSS. C_LI

neuroscience↗