Studying the brain from adolescence to adulthood through sparse multi-view matrix factorisations

Zi Wang, Vyacheslav Karolis, Chiara Nosarti, Giovanni Montana

Introduction

Research from King’s College London has introduced a sophisticated statistical framework to explore the structural divergence of male and female brains during maturation. Utilizing a novel matrix factorization technique, the study demonstrates that sexual dimorphism is not a static trait but a dynamic process that becomes significantly more pronounced between the ages of 15 and 30.

The Challenge of Longitudinal Brain Analysis

Understanding the structural "blueprints" of the brain is complicated by the fact that many cognitive and neuropsychiatric differences between sexes emerge or intensify during development. Traditional cross-sectional studies often fail to distinguish between static brain structures common to all individuals and age-specific variations that emerge at particular developmental stages. Standard tools like Principal Component Analysis (PCA) often struggle to isolate these shared versus age-specific factors across multiple groups.

Methodology: Multi-View Matrix Factorisation (MVMF)

To address these challenges, the researchers employed Multi-View Matrix Factorisation (MVMF), treating different age groups as distinct "views" of the data.

  • Variance Decomposition: The MVMF algorithm decomposes the total variance in Gray Matter Volume (GMV) into two independent components: a Shared Component (consistent across all ages) and an Age-Specific Component (unique to specific developmental windows).
  • Sparsity (sMVMF): By incorporating a sparsity penalty, the model automatically selects the most influential brain regions contributing to these differences, enhancing interpretability.
  • Visualization: The resulting factors allow for low-dimensional projections where individual subjects are represented as points, facilitating a visual assessment of how well the model distinguishes between sexes based on age-specific variance.

Datasets and Study Design

The study analyzed MRI data from two cohorts scanned at ages 15, 19, and 30:

  1. Control Cohort: Healthy individuals born at full-term.
  2. Preterm Cohort: Individuals born before 33 weeks of gestation, a group known for distinct neurodevelopmental trajectories.

The analysis focused on Gray Matter Volume (GMV) across 84 cortical and subcortical regions.

The Divergence Hypothesis

A central finding of the study is the increasing divergence of brain structure between sexes over time. At age 15, male and female brains appear structurally similar regarding age-specific variance. However, as individuals progress toward age 30, the data points for men and women become increasingly distinct. This suggests that neurodevelopmental trajectories for the sexes deviate significantly during the transition from adolescence to adulthood.

Visual separation of male and female cohorts across ages 15, 19, and 30, showing increasing divergence over time.

Key Results: Regional Specialization

The study identified specific brain regions that drive this divergence. By age 30, the regions most effectively distinguishing the sexes were primarily located within the limbic system, including the amygdala, hippocampus, and nucleus accumbens—areas critical for emotion, memory, and reward processing.

Ranking of brain regions by their contribution to sex-related variance at ages 19 and 30 in both preterm and control cohorts.

In the control cohort, age-specific factors accounted for approximately 9.4% of the variance at age 15, increasing to 14% by age 30. Interestingly, the preterm group showed earlier sex-based separation in executive function areas, such as the frontal pole, compared to the control group at age 19. The Temporo-parietal junction, involved in "Theory of Mind," also emerged as a top-ranking region for sex-related variance by age 30.

Limitations and Future Directions

While the study provides a clear map of where and when these structural changes occur, the underlying biological drivers—such as sex hormones or environmental influences—remain to be directly measured. Furthermore, while the method handles longitudinal data effectively, the subject composition across "views" was not perfectly identical.

Future applications of sMVMF could involve studying clinical populations to determine if atypical divergence in these age-specific factors can serve as a predictor for neuropsychiatric conditions like schizophrenia or depression, which frequently manifest during this critical developmental window.

Summary updated

BibTeX
@inproceedings{wang2016studying,
  author = {Zi Wang and Vyacheslav Karolis and Chiara Nosarti and Giovanni Montana},
  title = {{Studying the brain from adolescence to adulthood through sparse multi-view matrix factorisations}},
  booktitle = {PRNI},
  year = {2016},
  doi = {10.1109/prni.2016.7552340},
}
Figure