Linking Brain and Behavior

Linking Brain Measures to Phenotypes¶

In neuroscience, we believe that our brain data may contain some information about our phenotypes (i.e., any observable characteristics of an individual) such as age, cognition, behavior, or clinical status. But how do we study this?

In practice, this is usually studied in two related ways: association and prediction.

Association asks whether brain measures vary with a phenotype.

Prediction asks whether brain measures can estimate/predict that phenotype in new individuals.

1) Association: does the brain metric change with the phenotype?¶

Here the question is whether a brain feature differs across groups or varies along a continuous phenotype.

For example, in aging research, resting-state connectivity often shows reduced within-network connectivity and increased between-network connectivity with advancing age, suggesting a shift in network organization across the lifespan.

This can be expressed either as a group difference (“older adults vs younger adults”) or as a continuous relationship (“connectivity decreases/increases with age”).

This type of result is useful because it identifies which networks are most sensitive to the phenotype and helps generate mechanistic hypotheses about what is changing in the brain.

2) Prediction: can the brain metric estimate the phenotype in new people?¶

Here the goal is to train a model on brain features and test whether it can predict an unseen person’s phenotype. For instance, in our machine learning algorithm, the input would be some brain measures such as a functional connectivity matrix and the output can be the predicted outcome (e.g., young vs old, or actual age in years).

If the model can accurately predict the phenotype in new participants, it suggests that the brain measures contain meaningful information about that outcome. For example, if functional connectivity can predict age, then patterns of connectivity systematically change across the lifespan. Likewise, if brain measures can predict cognitive ability or symptom severity, it suggests that these traits are reflected in the brain’s organization.

Such models can be useful for identifying potential biomarkers for neuropsychiatric diseases, estimating an individual’s characteristics from brain data, and understanding which aspects of brain organization are most relevant to a given phenotype.

However, successful prediction alone does not imply that the brain measure causes the phenotype. It simply indicates that the two are related.