Summary: For years, neuroscientists have focused on the strongest 10% of brain signals, dismissing the rest as “noise.” However, a new study reveals that the other 90% of brain connections, the parts usually thrown away, can predict behavior with equal or even greater accuracy.

The research suggests that predictive information is widely distributed across the brain, meaning there isn’t just one “correct” network for a specific behavior, but many.

Key Findings

Multiple Pathways: The study proves that there are multiple, non-overlapping networks capable of predicting the same behavior. This suggests the brain has significant redundancy and “functional flexibility.”Psychiatric Implications: For conditions like depression, different individuals may rely on entirely different neural pathways to arrive at the same behavioral outcome.Therapeutic Targets: If several circuits can predict an illness, treatment shouldn’t be limited to the “top” networks. Targeting these overlooked circuits could provide a breakthrough for patients who are “treatment-resistant” to current therapies.The Accuracy Myth: High statistical strength does not necessarily mean higher biological relevance. The “noise” of today could be the precision medicine of tomorrow.

Source: Yale

Scientists who use imaging to understand the brain’s complexity often focus on the strongest signals and ignore the rest. But this strategy, researchers warn, may reveal only the tip of the iceberg.

A new study published in Nature Human Behavior reveals that connections routinely overlooked as “noise” during neuroimaging data analysis can predict behavior with remarkable accuracy—and implicate entirely different brain networks.

The finding could open many new targets for treating psychiatric illness, the researchers say.

“Many studies that rely on techniques like feature selection—which simplifies the brain down to a narrow slice—might only uncover a small part of the true neurobiology that underlies a given behavior,” says lead author Brendan Adkinson, PhD, an MD-PhD student at Yale School of Medicine.

“Our study suggests that there may be multiple, non-overlapping networks capable of predicting a given behavior just as well.”

Overlooked brain connections

One goal of human neuroimaging is to illuminate the brain mechanisms that drive cognition and mental health. But the complexity of brain connectivity makes data interpretation challenging. To address this, researchers often use feature selection, which focuses on the strongest 10% of brain connections to make the data easier to interpret.

For the study, researchers investigated whether signals discarded by feature selection could reveal meaningful insights about brain and behavior. The team examined brain imaging and behavioral data from more than 12,000 participants across four major U.S. datasets. For every participant, the team calculated the strength of association between brain connections and the outcome they wanted to predict.

All the connections were then ranked from the strongest to weakest associated and divided into 10 non-overlapping groups. Group one contained the top 10% of connections, those that scientists usually select, while groups two through 10 held the remaining 90% of connections—the connections often dismissed as noise. The team then built 10 prediction models, one for each group.

The team found that lower-ranked connections—groups two through nine—consistently achieved prediction accuracy similar to the top 10% of connections. In some cases, models built on lower groups of connections performed better than those trained on the top group. The authors suggest this might be because predictive information is widely distributed throughout brain connections and not just concentrated within the strongest ones.

“To our surprise, even when we completely excluded the networks people usually rely on to predict behavior, we still achieved nearly the same level of accuracy using everything that’s typically left behind,” says Adkinson, who works in the lab of senior author Dustin Scheinost, PhD, associate professor of radiology and biomedical imaging at YSM and associate director of biomedical imaging technology at the Yale Biomedical Imaging Institute.

Individual differences in mental health

The results indicate that by narrowing their focus, scientists risk oversimplifying the brain’s complexity, especially when dealing with brain disorders. For psychiatric disorders such as depression, individuals may rely on different neural pathways for the same behavior. And if several brain circuits can achieve similar prediction accuracy, it also suggests that therapeutic targets shouldn’t be limited to only the top networks.

“While the networks traditionally targeted by interventions may work for most patients, these overlooked networks might hold more utility for certain subsets of individuals,” says Adkinson.

“This could help explain why some people don’t currently respond to treatments that work for others.”

With these results, the team hopes to increase the clinical efficacy of brain-based biomarkers by better reflecting the brain’s complexity and individual variability.

Key Questions Answered:Q: If “weak” signals are so important, why did we ignore them for so long?

A: It’s a matter of data management. The brain has billions of connections. To make sense of the math, scientists “simplify” the brain by looking for the loudest voices in the room. This study shows that the “whispers” in the background are actually telling the same story, just in a different way.

Q: Does this mean current brain-based treatments are wrong?

A: Not wrong, just incomplete. Current treatments (like TMS or certain meds) target the “loudest” networks. This study explains why those treatments work for some but not others, some people’s brains might be using one of the “overlooked” networks instead.

Q: Can this help diagnose mental illness more accurately?

A: Yes. By including more of the brain’s complexity in our models, we can create better “biomarkers.” Instead of looking for one single “depression signal,” we can look at the whole “iceberg” to see which specific pathway is causing an issue for a specific person.

Editorial Notes:This article was edited by a Neuroscience News editor.Journal paper reviewed in full.Additional context added by our staff.About this mental health and neuroscience research news

Author: Colleen Moriarty
Source: Yale
Contact: Colleen Moriarty – Yale
Image: The image is credited to Neuroscience News

Original Research: Open access.
“Feature selection leads to divergent neurobiological interpretations of brain-based machine learning biomarkers” by Brendan D. Adkinson, Matthew Rosenblatt, Huili Sun, Javid Dadashkarimi, Link Tejavibulya, Corey Horien, Margaret L. Westwater, Raimundo X. Rodriguez, Stephanie Noble & Dustin Scheinost. Nature Human Behavior
DOI:10.1038/s41562-026-02447-y

Abstract

Feature selection leads to divergent neurobiological interpretations of brain-based machine learning biomarkers

A central objective in human neuroimaging is to understand the neurobiology underlying cognition and mental health.

Machine learning models trained on neuroimaging data are increasingly used as tools for predicting behavioural phenotypes, enhancing precision medicine and improving generalizability compared with traditional MRI studies.

However, the high dimensionality of brain connectivity data makes model interpretation challenging.

Prevailing practices rely on selecting features and, implicitly, interpreting identified feature networks as uniquely representative of a given phenotype while overlooking others.

Despite its widespread use, how univariate feature selection balances the trade-off between simplification for optimizing modelling and oversimplification that misrepresents true neurobiology remains understudied.

Here, using four large-scale neuroimaging datasets spanning over 12,000 participants and 13 outcomes, we demonstrate that edges discarded by feature selection can achieve significant prediction accuracies while yielding different neurobiological interpretations.

These results are observed across cognitive, developmental and psychiatric phenotypes, extend to both functional connectivity (functional MRI) and structural (diffusion tensor imaging) connectomes, and remain evident in external validation.

They suggest that focusing on only the top features may simplify the neurobiological bases of brain–behaviour associations.

Such interpretations present only the tip of the iceberg when certain disregarded features may be just as meaningful, potentially contributing to ongoing issues surrounding reproducibility within the field.

More broadly, our results reinforce that subtle brain-wide signals should not be ignored.