Connectome generative models, otherwise known as generative network models, provide insight into the wiring principles underpinning brain network organization. While these models can approximate numerous statistical properties of empirical networks, they typically fail to explicitly characterize an important contributor to brain organization—axonal growth. Emulating the chemoaffinity-guided axonal growth, we provide a novel generative model in which axons dynamically steer the direction of propagation based on distance-dependent chemoattractive forces acting on their growth cones. This simple dynamic growth mechanism, despite being solely geometry-dependent, is shown to generate axonal fiber bundles with brain-like geometry and features of complex network architecture consistent with the human brain, including lognormally distributed connectivity weights, scale-free nodal degrees, small-worldness, and modularity. We demonstrate that our model parameters can be fitted to individual connectomes, enabling connectome dimensionality reduction and comparison of parameters between groups. Our work offers an opportunity to bridge studies of axon guidance and connectome development, providing new avenues for understanding neural development from a computational perspective. Author Summary Generative models of the human connectome provide insight into principles driving brain network development. However, current models do not capture axonal outgrowth, which is crucial to the formation of neural circuits. We develop a novel generative connectome model featuring dynamic axonal outgrowth, revealing the contribution of microscopic axonal guidance to the network topology and axonal geometry of macroscopic connectomes. Simple axonal outgrowth rules representing continuous chemoaffinity gradients are shown to generate complex, brain-like topologies and realistic axonal fascicle architectures. Our model is sufficiently sensitive to capture subtle interindividual differences in axonal outgrowth between healthy adults. Our results are significant because they reveal core principles that may give rise to both complex brain networks and brain-like axonal bundles, unifying neurogenesis across scales.

Several recent studies have optimized deep neural networks to learn high-dimensional relationships linking structural and functional connectivity across the human connectome. However, the extent to which these models recapitulate individual-specific characteristics of resting-state functional brain networks remains unclear. A core concern relates to whether current individual predictions outperform simple benchmarks such as group averages and null conditions. Here, we consider two measures to statistically evaluate whether functional connectivity predictions capture individual effects. We revisit our previously published functional connectivity predictions for 1,000 healthy adults and provide multiple lines of evidence supporting that our predictions successfully capture subtle individual-specific variation in connectivity. While predicted individual effects are statistically significant and outperform several benchmarks, we find that effect sizes are small (i.e., 8%–11% improvement relative to group-average benchmarks). As such, initial expectations about individual prediction performance expressed by us and others may require moderation. We conclude that individual predictions can significantly outperform appropriate benchmark conditions and we provide several recommendations for future studies in this area. Future studies should statistically assess the individual prediction performance of their models using one of the measures and benchmarks provided here. Author Summary Functional and structural brain networks share considerable overlap in network architecture. However, it remains debated whether deep neural networks can be trained to predict an individual's functional brain network from their structural connectome. We demonstrate that individual variability in functional brain connectivity can be successfully predicted from an individual's connectome, although prediction performance is modest when benchmarked against appropriate null models. We provide recommendations for future studies aiming to evaluate such predictions, specifically considering the impact of Riemann geometry, adjustments for cross-validation induced dependence and standardization. Accurate prediction models will enable extrapolation of functional networks for individuals without empirically acquired functional MRI data or noisy data. They may also facilitate digital simulations of the potential functional consequences arising from pathological changes in the connectome.