The neural basis of intelligence in fine-grained cortical topographies

  1. Ma Feilong
  2. J. Swaroop Guntupalli
  3. James V. Haxby

  • Curated by eLife

    eLife logo

    Summary: In this work, Feilong and colleagues use the Human Connectome Project fMRI data to investigate the degree to which the strength of functional connectivity is predictive of general intelligence, and the degree to which that predictive power is improved using the hyperalignment procedures their lab has developed. More specifically, the authors predict general intelligence using either coarse-grained functional connectivity (based on 360 ROIs) or fine-grained functional connectivity (vertex-wise) after hyperalignment. The results show a two-fold increase in variance explained in general intelligence between coarse-grained and fine-grained connectivity. This is a very clearly-written paper that presents an important result, which has the potential of great impact on the field of behavioral prediction. However, the reviewers and editors do have some significant concerns with the predictive modeling presented in this work.

    Reviewer #2 and Reviewer #3 opted to reveal their name to the authors in the decision letter after review.

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

Intelligent thought is the product of efficient neural information processing, which is embedded in fine-grained, topographically-organized population responses and supported by fine-grained patterns of connectivity among cortical fields. Previous work on the neural basis of intelligence, however, has focused on coarse-grained features of brain anatomy and function, because cortical topographies are highly idiosyncratic at a finer scale, obscuring individual differences in fine-grained connectivity patterns. We used a computational algorithm, hyperalignment, to resolve these topographic idiosyncrasies, and found that predictions of general intelligence based on fine-grained (vertex-by-vertex) connectivity patterns were markedly stronger than predictions based on coarse-grained (region-by-region) patterns. Intelligence was best predicted by fine-grained connectivity in the default and frontoparietal cortical systems, both of which are associated with self-generated thought. Previous work overlooked fine-grained architecture because existing methods couldn’t resolve idiosyncratic topographies, preventing investigation where the keys to the neural basis of intelligence are more likely to be found.

Article activity feed

  1. eLife

    Reviewer #3:

    In this work, Feilong and colleagues use Human Connectome Project fMRI data to investigate the degree to which the strength of functional connectivity is predictive of general intelligence, and the degree to which that predictive power is improved using the hyperalignment procedures their lab has previously developed. I am broadly very supportive of the goals of improving prediction of individual behavioral differences via improved, functionally-based cross-subject registration, and I have always felt that the hyperalignment procedure is one of the most promising approaches for improving cross-subject functional registration. Overall I feel that this paper is an important next step in the development and maturation of the hyperalignment technique.

    However, I do have two significant concerns with the predictive modeling presented in this work. I note that I am not an expert in these techniques, so these concerns may be due to my own ignorance; however, I would like to see the authors at least better explain these issues to non-experts like myself.

    First, the authors employed a leave-one-family-out cross-validation scheme for their predictive modeling. My understanding is that the field has generally moved away from leave-one-out or leave-few-out cross-validation, as that approach consistently overestimates the predictive power of generated models. The HCP is a large dataset. Can the authors employ a more robust approach of using fully split halves?

    Second, the authors make the claim that fine-grained (vertex-wise) connectivity has substantially better predictive power than coarse-grained (parcel-wise) connectivity, based on the variance in intelligence explained by the predictive models. However, the models based on fine-grained connectivity also have many, many more variables being used to make the prediction. Is this not a confound?

  2. eLife

    Reviewer #2:

    Summary:

    This paper predicts intelligence using either coarse-grained functional connectivity (based on 360 ROIs) or fine-grained functional connectivity (vertex-wise) after hyperalignment. The results show a two-fold increase of variance explained in general intelligence between coarse-grained and fine-grained connectivity.

    General:

    This is a very clearly-written paper that presents an important result, which has the potential of great impact on the field of behavioral prediction. My comments below are relatively minor and primarily aimed at clarifying a few details in the article. Please find my detailed comments below, approximately in order of importance.

    Major comments:

    1. The fine-grained functional connectivity has richer features than coarse-grained, leading to higher dimensionality in the PCA step (supplementary figure S5). I wonder if this might contribute to improved prediction accuracy. Related to this, it appears that there may also be a relationship between PCA dimensionality and regularization parameter, such that more regularization may be needed when more PCs are used in the model. It would be interesting to test the effect of fixing the PCA dimensionality (and perhaps also the regularization) across all models to control model complexity.

    2. The Glasser 360 parcellation was used throughout this work. There are subject-specific parcels and group-level parcels available for this parcellation. Please clarify which of these were used. If the group-level parcels were used, it might be interesting to see how the coarse-grained prediction accuracies might improve when using subject-specific parcels.

    3. The residuals of fine-grained connectivity profiles were obtained after subtracting coarse-grain connectivity. Why was subtraction used here, rather than regressing out (i.e., orthogonalizing with respect to) the coarse-grained connectivity?

  3. eLife

    Reviewer #1:

    In this study, Feilong and colleagues showed that hyper-aligned fine-grained cortical connectivity profiles can be used to strongly predict general intelligence in individual participants. This is an important study demonstrating the utility of previously developed connectivity hyperalignment and highlighting the behavioral importance of fine-grained connectivity which is typically ignored in more standard functional connectivity analysis.

    1. How does the bootstrapping handle the family structure in the data? More details are needed.

    2. The authors mentioned that "the code for performing hyperalignment and nuisance regression was adapted from PyMVPA". One of the most important contributions of this study is the impressive demonstration of prediction performance improvement using hyperalignment and fine-grained connectivity profiles. Therefore, it is important that the adapted code and code utilized for the current study be made publicly available. While connectivity hyperalignment code from the previous study is available in PyMVPA, my experience is that it is not easy to use. If no code from the current study is made available, I believe it will be very difficult to replicate this study.

  4. eLife

    Summary: In this work, Feilong and colleagues use the Human Connectome Project fMRI data to investigate the degree to which the strength of functional connectivity is predictive of general intelligence, and the degree to which that predictive power is improved using the hyperalignment procedures their lab has developed. More specifically, the authors predict general intelligence using either coarse-grained functional connectivity (based on 360 ROIs) or fine-grained functional connectivity (vertex-wise) after hyperalignment. The results show a two-fold increase in variance explained in general intelligence between coarse-grained and fine-grained connectivity. This is a very clearly-written paper that presents an important result, which has the potential of great impact on the field of behavioral prediction. However, the reviewers and editors do have some significant concerns with the predictive modeling presented in this work.

    Reviewer #2 and Reviewer #3 opted to reveal their name to the authors in the decision letter after review.

  5. Version published to 10.1101/2020.06.06.138099 on bioRxiv