Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 52 of the 52 references it cites were found in a registry or an index.

Connectotyping: Model Based Fingerprinting of the Functional Connectome

Open access

Abstract

A better characterization of how an individual's brain is functionally organized will likely bring dramatic advances to many fields of study. Here we show a model-based approach toward characterizing resting state functional connectivity MRI (rs-fcMRI) that is capable of identifying a so-called "connectotype", or functional fingerprint in individual participants. The approach rests on a simple linear model that proposes the activity of a given brain region can be described by the weighted sum of its functional neighboring regions. The resulting coefficients correspond to a personalized model-based connectivity matrix that is capable of predicting the timeseries of each subject. Importantly, the model itself is subject specific and has the ability to predict an individual at a later date using a limited number of non-sequential frames. While we show that there is a significant amount of shared variance between models across subjects, the model's ability to discriminate an individual is driven by unique connections in higher order control regions in frontal and parietal cortices. Furthermore, we show that the connectotype is present in non-human primates as well, highlighting the translational potential of the approach.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1371/journal.pone.0111048

Cites

52 of 52 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Individualized Markov ROIs. For each T1-weighted MP-RAGE volume, we individually segmented each brain after individual registration to the fsaverage surface using Freesurfer. An individual model was calculated in the 20 subjects of whom individualized ROI generation was possible (see methods).
Figure 1 Individualized Markov ROIs. For each T1-weighted MP-RAGE volume, we individually segmented each brain after individual registration to the fsaverage surface using Freesurfer. An individual model was calculated in the 20 subjects of whom individualized ROI generation was possible (see methods).CC BY · PubMed Central
Figure 2 Connectotyping: Once the time courses are created for each ROI (panel A1), a mathematical model is fit to predict the activity of each ROI given its historical values (panel A2). Then, the estimations are subtracted from the original time courses. Next, each ROI’s residual is modeled as the weighted sum of the residuals of all other ROIs (panel A3). Finally, all the coefficients are grouped in a matrix, or “the model”. Each subject’s timecourses can be used to make an estimation of the actual values (panel B1) and the fit can be quantified by the correlation coefficient of measured and estimated timecourses. This approach allows us to test the model in the same subject the model was obtained from or using a different subject (panel B2).
Figure 2 Connectotyping: Once the time courses are created for each ROI (panel A1), a mathematical model is fit to predict the activity of each ROI given its historical values (panel A2). Then, the estimations are subtracted from the original time courses. Next, each ROI’s residual is modeled as the weighted sum of the residuals of all other ROIs (panel A3). Finally, all the coefficients are grouped in a matrix, or “the model”. Each subject’s timecourses can be used to make an estimation of the actual values (panel B1) and the fit can be quantified by the correlation coefficient of measured and estimated timecourses. This approach allows us to test the model in the same subject the model was obtained from or using a different subject (panel B2).CC BY · PubMed Central
Figure 3 Modeling results. Panel A shows the measured and predicted timecourse using both the pinv and tsvd methods of the ROI L-V1 for the first subject of group H27 (20 frames are shown from the Markov parcellation). Panel B shows the distribution of the correlation coefficients between the predicted and measured ROIs calculated by the pinv and tsvd methods, respectively. Panels C shows the distribution of correlation coefficients between predictions and measured time courses for 27 subjects, using the FVE and Markov parcellations, pinv and tsvd-based predictions are shown on each panel. Thin lines indicate the range of values, thick lines correspond to the 25 to 75 percentiles and the vertical marker is the median of each distribution. Both methods show a strong ability to predict the timeseries of a given ROI within an individual.
Figure 3 Modeling results. Panel A shows the measured and predicted timecourse using both the pinv and tsvd methods of the ROI L-V1 for the first subject of group H27 (20 frames are shown from the Markov parcellation). Panel B shows the distribution of the correlation coefficients between the predicted and measured ROIs calculated by the pinv and tsvd methods, respectively. Panels C shows the distribution of correlation coefficients between predictions and measured time courses for 27 subjects, using the FVE and Markov parcellations, pinv and tsvd-based predictions are shown on each panel. Thin lines indicate the range of values, thick lines correspond to the 25 to 75 percentiles and the vertical marker is the median of each distribution. Both methods show a strong ability to predict the timeseries of a given ROI within an individual.CC BY · PubMed Central
Figure 4 Predicting subjects in group H5. Model (pinv and tsvd) predicting same subject on second scan date for group H5 (i.e. predicting itself versus predicting others) for the FVE and Markov parcellations. Case "Itself" corresponds to the result when the model, regularization, and prediction were performed on the same subject. "Itself reg. in others" is the case when model and prediction were performed in the same subject, but regularization was performed in a different subject. "Fooled by regularization" comes from the comparison where the regularization and prediction were performed in the same subject, but the model was obtained from a different subject. The remaining combinations are concatenated in the case "Others". In all instances, the model performs more strongly in predicting self versus others; however, a significant amount of variance is accounted for in the “other” cases.
Figure 4 Predicting subjects in group H5. Model (pinv and tsvd) predicting same subject on second scan date for group H5 (i.e. predicting itself versus predicting others) for the FVE and Markov parcellations. Case "Itself" corresponds to the result when the model, regularization, and prediction were performed on the same subject. "Itself reg. in others" is the case when model and prediction were performed in the same subject, but regularization was performed in a different subject. "Fooled by regularization" comes from the comparison where the regularization and prediction were performed in the same subject, but the model was obtained from a different subject. The remaining combinations are concatenated in the case "Others". In all instances, the model performs more strongly in predicting self versus others; however, a significant amount of variance is accounted for in the “other” cases.CC BY · PubMed Central
Figure 5 Classification power quantified by an ROC curve (using tsvd), group H5. The left panel shows the distributions (top) and the corresponding histograms (bottom) of the correlations of each subject predicting itself (Pos) and predicting others (Neg). The colored line is the cumulative distribution of correlations of the entire data set. The optimal classification is achieved when the threshold is set to 0.7825, which corresponds to the 5.83% of the largest correlation coefficients. This threshold leads to a true positive rate (TPR) of 1 and a false positive rate (FPR) of 0. As shown the classification accuracy is 100% with marked distinctions in the distributions. The performance here using tsvd was stronger than that of the pinv (Figure S4 in File S1). Analyses were performed using the Markov atlas.
Figure 5 Classification power quantified by an ROC curve (using tsvd), group H5. The left panel shows the distributions (top) and the corresponding histograms (bottom) of the correlations of each subject predicting itself (Pos) and predicting others (Neg). The colored line is the cumulative distribution of correlations of the entire data set. The optimal classification is achieved when the threshold is set to 0.7825, which corresponds to the 5.83% of the largest correlation coefficients. This threshold leads to a true positive rate (TPR) of 1 and a false positive rate (FPR) of 0. As shown the classification accuracy is 100% with marked distinctions in the distributions. The performance here using tsvd was stronger than that of the pinv (Figure S4 in File S1). Analyses were performed using the Markov atlas.CC BY · PubMed Central
Figure 6 Predictions using individualized ROIs. A) This panel shows the average correlation coefficient when predicted and measured BOLD activity is compared. On the left, the results using the pinv method is used to calculate the model using 45% of the frames and predicting 15% of fresh in-sample data. On the right, indicates the 4 distinct groups when the tsvd method is used to predict data using 45% of frames to obtain a model, 15% of fresh in sample-data to select the number of singular values to preserve, and 15% of fresh in-sample data to predict bold activity according to the regularized model. B) This panel summarizes the predictive power using receiver operating characteristic curves (here, the singular values where calculated by finding stable norms of residuals and estimations, as described in methods).
Figure 6 Predictions using individualized ROIs. A) This panel shows the average correlation coefficient when predicted and measured BOLD activity is compared. On the left, the results using the pinv method is used to calculate the model using 45% of the frames and predicting 15% of fresh in-sample data. On the right, indicates the 4 distinct groups when the tsvd method is used to predict data using 45% of frames to obtain a model, 15% of fresh in sample-data to select the number of singular values to preserve, and 15% of fresh in-sample data to predict bold activity according to the regularized model. B) This panel summarizes the predictive power using receiver operating characteristic curves (here, the singular values where calculated by finding stable norms of residuals and estimations, as described in methods).CC BY · PubMed Central
Figure 7 Predicting subjects in group M11 (11 Macaques). Distribution of average correlation coefficients for group M11 predicting itself versus predicting others. For tsvd, case 1 (itself): model, regularization, and prediction were performed in the same subject. Case 2 (itself regularized in others): model and prediction were performed in the same subject, but regularization (4) was performed in a different subject. Case 3 (fooled by regularization): the regularization and prediction were performed in the same subject, but the model was obtained from a different subject. Case 4 (others): all the remaining possibilities.
Figure 7 Predicting subjects in group M11 (11 Macaques). Distribution of average correlation coefficients for group M11 predicting itself versus predicting others. For tsvd, case 1 (itself): model, regularization, and prediction were performed in the same subject. Case 2 (itself regularized in others): model and prediction were performed in the same subject, but regularization (4) was performed in a different subject. Case 3 (fooled by regularization): the regularization and prediction were performed in the same subject, but the model was obtained from a different subject. Case 4 (others): all the remaining possibilities.CC BY · PubMed Central
Figure 8 Predictions in the group of 5 human subjects (H5– second day scan) based on consistent or unique connection weights across the populations. After using tsvd to generate a model for each subject, the variance across participants was calculated for each entry (panel A). Panels B and C shows the distribution and cumulative distribution of the coefficient’s variance, respectively. From the cumulative distribution, the connections were grouped into 100 bins with similar variance (presented in panel A). Then, 100 masks were created setting to zero the connections with higher variance and incrementally using more preserved (lower variance) connections. The masks were applied to the first day model to predict the second day scan. Panel D shows the correlation coefficient between the predicted and actual time courses for each one of the subjects in the group H5. The cell i, j indicates that the model was obtained in the subject’s i first-day scan and that model was used to predict the subject’s j second-day time-courses. The sub-panel’s title indicates the percentile of most preserved connections (mask) used in the model. Panel E shows the 100 ROC (lines) and optimal threshold (dots) for each run (percentile). Panel F shows the optimal True Positive Rate (TPR) and False Positive Rate (FPR) as a function of the run. Panel G shows the distributions of correlation coefficients for all the subjects in group H5 predicting itself and predicting others, as a function of run. The dotted line indicates the point at which there is a clear separation between the two predictions. All of panels D-F highlight how the predictive nature of the model is dependent on the most variable connections across subjects. The most preserved (lowest variance) connections do little to distinguish between individuals.
Figure 8 Predictions in the group of 5 human subjects (H5– second day scan) based on consistent or unique connection weights across the populations. After using tsvd to generate a model for each subject, the variance across participants was calculated for each entry (panel A). Panels B and C shows the distribution and cumulative distribution of the coefficient’s variance, respectively. From the cumulative distribution, the connections were grouped into 100 bins with similar variance (presented in panel A). Then, 100 masks were created setting to zero the connections with higher variance and incrementally using more preserved (lower variance) connections. The masks were applied to the first day model to predict the second day scan. Panel D shows the correlation coefficient between the predicted and actual time courses for each one of the subjects in the group H5. The cell i, j indicates that the model was obtained in the subject’s i first-day scan and that model was used to predict the subject’s j second-day time-courses. The sub-panel’s title indicates the percentile of most preserved connections (mask) used in the model. Panel E shows the 100 ROC (lines) and optimal threshold (dots) for each run (percentile). Panel F shows the optimal True Positive Rate (TPR) and False Positive Rate (FPR) as a function of the run. Panel G shows the distributions of correlation coefficients for all the subjects in group H5 predicting itself and predicting others, as a function of run. The dotted line indicates the point at which there is a clear separation between the two predictions. All of panels D-F highlight how the predictive nature of the model is dependent on the most variable connections across subjects. The most preserved (lowest variance) connections do little to distinguish between individuals.CC BY · PubMed Central
Figure 9 Shared vs individualized connections. The color indicates how stable the source connections are for each colored ROI. Color (in base 10 logarithmic scale) indicates the sum of the variance of the coefficients. The lower the value, the more stable (i.e., the more preserved) the connection across individuals (shown in the Markov parcellation).
Figure 9 Shared vs individualized connections. The color indicates how stable the source connections are for each colored ROI. Color (in base 10 logarithmic scale) indicates the sum of the variance of the coefficients. The lower the value, the more stable (i.e., the more preserved) the connection across individuals (shown in the Markov parcellation).CC BY · PubMed Central
Figure 10 Frames required for prediction. The distribution of correlation coefficients for each model predicting itself (green) vs predicting others (red) is shown as a function of frames for the group H27. 60 frames produces stable self prediction, where below 60 frames the model is slightly reduced in its ability to predict an individual.
Figure 10 Frames required for prediction. The distribution of correlation coefficients for each model predicting itself (green) vs predicting others (red) is shown as a function of frames for the group H27. 60 frames produces stable self prediction, where below 60 frames the model is slightly reduced in its ability to predict an individual.CC BY · PubMed Central
Figure 11 Group H27 resting state networks for 20, 60 and the entire number of frames used to calculate the model based tsvd connectivity map on the Markov atlas. A connection density of 20% (i.e. strongest 20% of coefficients) was used. Seed based analysis shows that fronto-parietal and default mode networks are well represented at the group level even down to 20 frames. Similar phenomena can be demonstrated in an individual subject (Figure S8 in File S1).
Figure 11 Group H27 resting state networks for 20, 60 and the entire number of frames used to calculate the model based tsvd connectivity map on the Markov atlas. A connection density of 20% (i.e. strongest 20% of coefficients) was used. Seed based analysis shows that fronto-parietal and default mode networks are well represented at the group level even down to 20 frames. Similar phenomena can be demonstrated in an individual subject (Figure S8 in File S1).CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 52 references

Cited by

See all 203

Metadata

Authors
Miranda-Dominguez, O., Mills, B. D., Carpenter, S. D., Grant, K. A., Kroenke, C. D., Nigg, J. T., & Fair, D. A.
Affiliations
  • Oregon Health & Science University (US)
  • Oregon National Primate Research Center (US)
Corresponding author
Fair, Damien A.
Venue
PLoS ONE
Published
2014-11-11
Volume
9
Issue
11
Pages
e111048
ISSN
1932-6203 (ISSN-L)
OpenAlex
W2058587070
Cited by
197
References
52
Citations by year
2026: 6 · 2025: 14 · 2024: 12 · 2023: 19 · 2022: 28 · 2021: 29 · 2020: 19 · 2019: 25 · 2018: 20 · 2017: 17 · 2016: 6 · 2015: 2
FWCI
4.65
Type
journal article
Language
en
Open access
Locations
  • PLoS ONE (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • James S. McDonnell Foundation (R01 MH096773, K99/R00 MH091238)
  • National Institutes of Health (P51OD011092)
  • Oregon Clinical and Translational Research Institute
  • National Institute on Alcohol Abuse and Alcoholism (AA 109431, AA13510, U01AA013510)
  • National Institute of Mental Health (K99/R00 MH091238, K99MH091238, 1R01MH86654, R00MH091238, R01MH096773, R01MH086654)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Functional Brain Connectivity Studies
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Advanced MRI Techniques and Applications
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Advanced Neuroimaging Techniques and Applications
Keywords
functional connectome, resting-state functional connectivity, individual differences, brain network topology, frontal and parietal cortex, non-human primates, time series forecasting
MeSH terms
Adult, Algorithms, Animals, Female, Frontal Lobe, Humans, Image Processing, Computer-Assisted, Macaca fascicularis, Magnetic Resonance Imaging, Male, Models, Theoretical, Nerve Net, Neural Pathways, Parietal Lobe, Young Adult, Connectome

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Connectotyping: Model Based Fingerprinting of the Functional Connectome. https://app.cochat.ai/papers/10.1371/journal.pone.0111048
The paper
Miranda-Dominguez, O., Mills, B. D., Carpenter, S. D., Grant, K. A., Kroenke, C. D., Nigg, J. T., & Fair, D. A. (2014). Connectotyping: Model Based Fingerprinting of the Functional Connectome. PLoS ONE. https://doi.org/10.1371/journal.pone.0111048

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.