Boosting hyperalignment performance with age-specific templates
- Dartmouth College, United States
- University of Bologna, Italy
- University of South Carolina, United States
Figures
Figure 1
Inter-subject correlation (ISC) results (Cam-CAN).
(a) The average z (± SD) for the old group was 0.6897 ± 0.0759 for congruent templates, and 0.6700 ± 0.0839 for incongruent templates. The mean ISC difference (± SE) was 0.0197 ± 0.0017, t(212) = 11.33, Cohen’s d = 0.7766, and p < 10–22. The average z (± SD) for the young group was 0.8566 ± 0.1151 for congruent templates, and 0.8337 ± 0.1069 for incongruent templates. The mean ISC difference (± SE) was 0.0229 ± 0.0008, t(209) = 30.09, Cohen’s d = 2.0765, and p < 10–77. (b) Scatter plot of each participant’s average ISC value derived from different congruent and incongruent templates. (c) Topographic distribution of ISC difference across different brain regions.
-
Figure 1—source data 1
Source data for Inter-subject correlation (ISC) results (Cam-CAN).
- https://cdn.elifesciences.org/articles/110566/elife-110566-fig1-data1-v2.zip
Figure 2
Prediction performance comparison between congruent and incongruent templates.
(a) The average z (± SD) for the old group was 0.4279 ± 0.0558 for congruent templates, and 0.4126 ± 0.0597 for incongruent templates. The mean difference of predicted connectome (± SE) was 0.0154 ± 0.0007, t(212) = 21.91, Cohen’s d = 1.5014, and p < 10–55. The average z (± SD) for the young group was 0.5716 ± 0.0779 for congruent templates, and 0.5568 ± 0.0743 for incongruent templates. The mean difference of predicted connectome (± SE) was 0.0147±0.0004, t(209) = 39.07, Cohen’s d = 2.6963, and p < 10–97. (b) Scatter plot of each participant’s average correlation value derived from different congruent and incongruent templates. (c) Topographic correlation difference across different brain regions.
-
Figure 2—source data 1
Source data for prediction performance comparison between congruent and incongruent templates (Cam-CAN).
- https://cdn.elifesciences.org/articles/110566/elife-110566-fig2-data1-v2.zip
Figure 3
Comparison of individual connectome prediction across all age spans.
(a) Scatter plot showing Pearson correlation between actual connectome and predicted connectome derived from young and old templates for individuals across all age spans. (b) Correlation difference among individuals between two age group templates, ordered by age.
-
Figure 3—source data 1
Source data for comparison of individual connectome prediction across all age spans (Cam-CAN).
- https://cdn.elifesciences.org/articles/110566/elife-110566-fig3-data1-v2.zip
Figure 4
Predicting brain responses to the movie based on congruent and incongruent templates.
(a) The average z (± SD) for the old group was 0.1661 ± 0.0600 for congruent templates, and 0.1612±0.0636 for incongruent templates. The mean difference (± SE) was 0.0049 ± 0.0010, t(212) = 5.02, Cohen’s d = 0.3443, and p < 10–5. The average z (± SD) for the young group was 0.2925 ± 0.0627 for congruent templates, and 0.2441 ± 0.0490 for incongruent templates. The mean difference (± SE) was 0.0485 ± 0.0013, t(209) = 35.99, Cohen’s d = 2.4833, and p < 10–90. (b) Scatter plot of each participant’s average predicted response values derived from different congruent and incongruent templates. (c) Topographic distribution of predicted response difference across different brain regions.
-
Figure 4—source data 1
Source data for predicting brain responses to the movie based on congruent and incongruent templates (Cam-CAN).
- https://cdn.elifesciences.org/articles/110566/elife-110566-fig4-data1-v2.zip
Figure 5
Schematic of the procedure for building and testing hyperalignment templates.
Darker arrows indicate congruent templates (i.e., from the same age group), whereas light arrows indicate incongruent templates.
Appendix 1—figure 1
Schematic of the procedure for building and testing hyperalignment templates.
Appendix 1—figure 2
tSNR differences between age groups.
(a) The distribution of tSNR across participants, separately for each age group. The distribution of tSNR across the cortex for the old group (b) and the young group (c).
-
Appendix 1—figure 2—source data 1
Source data for tSNR differences between age groups.
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig2-data1-v2.zip
Appendix 1—figure 3
Inter-subject correlation results (DLBS).
(a) The average z (± SD) for the old group was 0.6044 ± 0.0789 for congruent templates, and 0.6020 ± 0.0810 for incongruent templates. The mean ISC difference (± SE) was 0.0025 ± 0.0006, t(181) = 4.16, Cohen’s d = 0.3087, and p < 10–4. The average z (± SD) for the young group was 0.7087 ± 0.0827 for congruent templates, and 0.7034 ± 0.0809 for incongruent templates. The mean ISC difference (± SE) was 0.0053 ± 0.0004, t(91) = 13.24, Cohen’s d = 1.3806, and p < 10–22. (b) Scatter plot of each participant’s average ISC value derived from different congruent and incongruent templates. (c) Topographic distribution of ISC difference across different brain regions.
-
Appendix 1—figure 3—source data 1
Source data for inter-subject correlation results (DLBS).
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig3-data1-v2.zip
Appendix 1—figure 4
Inter-subject correlation results based on the middle-aged cohort (Cam-CAN).
(a) Mean ISC difference (Congruent − Incongruent) for the middle-aged and young groups. The mean ISC difference (± SE) for the middle-aged group was −0.0009 ± 0.0005, while the mean ISC difference for the young group was 0.0064 ± 0.0003. (b) Scatter plot of each participant’s average ISC value derived from congruent and incongruent templates for the middle-aged versus young comparison. (c) Mean ISC difference (Congruent − Incongruent) for the old- and middle-aged groups. The mean ISC difference (± SE) for the old group was 0.0079 ± 0.0009, and the mean ISC difference for the middle-aged group was 0.0088 ± 0.0007. (d) Scatter plot of each participant’s average ISC value derived from congruent and incongruent templates for the old- versus middle-aged comparison.
-
Appendix 1—figure 4—source data 1
Source data for inter-subject correlation results based on the middle-aged cohort (Cam-CAN).
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig4-data1-v2.zip
Appendix 1—figure 5
Connectome prediction results based on the middle-aged cohort (Cam-CAN).
(a) Mean z-difference (Congruent − Incongruent) in connectome prediction accuracy for the middle-aged and young groups. The mean difference (± SE) for the middle-aged group was 0.0014 ± 0.0003, while the mean difference for the young group was 0.0034 ± 0.0002. (b) Scatter plot of each participant’s connectome prediction correlation derived from congruent and incongruent templates for the middle-aged versus young comparison. (c) Mean z-difference (Congruent − Incongruent) in connectome prediction accuracy for the old- and middle-aged groups. The mean difference (± SE) for the old group was 0.0056 ± 0.0004, and the mean difference for the middle-aged group was 0.0051 ± 0.0002. (d) Scatter plot of each participant’s connectome prediction correlation derived from congruent and incongruent templates for the old- versus middle-aged comparison.
-
Appendix 1—figure 5—source data 1
Source data for connectome prediction results based on the middle-aged cohort (Cam-CAN).
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig5-data1-v2.zip
Appendix 1—figure 6
Connectome prediction results with 10-year age templates: (a) Mean Fisher’s z-difference (Congruent − Incongruent) in connectome prediction for the young cohort (20–30 years old).
The bar plot compares the performance of the age-congruent template against incongruent templates from progressively older cohorts (ranging from 30–40 to 80–90). (b) Mean Fisher’s z-difference (Congruent − Incongruent) in connectome prediction for the older cohort (80–90 years old), comparing the age-congruent template against incongruent templates from progressively younger cohorts (ranging from 20–30 to 70–80).
-
Appendix 1—figure 6—source data 1
Source data for connectome prediction results with 10-year age templates.
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig6-data1-v2.zip
Tables
Table 1
Age group division and corresponding training/test set sizes.
| Age group | Age range | Number of participants | Training group | Test group |
|---|---|---|---|---|
| Young | 18–45 | 215 | 144 | 71 |
| Old | 65–90 | 216 | 144 | 72 |
Additional files
-
MDAR checklist
- https://cdn.elifesciences.org/articles/110566/elife-110566-mdarchecklist1-v2.docx
-
Supplementary file 1
All subgraphs regenerated using the minimal data files.
- https://cdn.elifesciences.org/articles/110566/elife-110566-supp1-v2.zip
-
Source data 1
Code to regenerate all graphs from the data files provided.
- https://cdn.elifesciences.org/articles/110566/elife-110566-data1-v2.zip
-
Appendix 1—figure 2—source data 1
Source data for tSNR differences between age groups.
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig2-data1-v2.zip
-
Appendix 1—figure 3—source data 1
Source data for inter-subject correlation results (DLBS).
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig3-data1-v2.zip
-
Appendix 1—figure 4—source data 1
Source data for inter-subject correlation results based on the middle-aged cohort (Cam-CAN).
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig4-data1-v2.zip
-
Appendix 1—figure 5—source data 1
Source data for connectome prediction results based on the middle-aged cohort (Cam-CAN).
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig5-data1-v2.zip
-
Appendix 1—figure 6—source data 1
Source data for connectome prediction results with 10-year age templates.
- https://cdn.elifesciences.org/articles/110566/elife-110566-app1-fig6-data1-v2.zip
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Boosting hyperalignment performance with age-specific templates
eLife 15:RP110566.
https://doi.org/10.7554/eLife.110566.3
