Dyads, typicality, and alignment: what ISC and IS-RSA can actually attribute

Question: How do inter-subject correlation and inter-subject representational similarity analysis of naturalistic fMRI distinguish shared stimulus-driven responses from stable individual differences in brain activity?

Answer

Across the retrieved papers, neither ISC nor IS-RSA performs this separation on its own: ISC quantifies how much of a response is shared along the stimulus time course but is silent about why it is shared, and IS-RSA relocates the question to participant pairs, testing whether pairwise neural similarity covaries with pairwise trait or behaviour similarity. What the corpus collectively shows is that the inferential work is done by three things that surround the statistic: stimulus design that makes content trait-relevant or subtracts a control condition, functional alignment that removes topographical idiosyncrasy before similarity is computed, and independent evidence that the individual-difference component is stable (prospective prediction, within-person stability, or heritability). A striking convergence is that trait-linked structure in these data very often appears as deviation from consensus—lower ISC, lower typicality, greater heterogeneity—rather than as a positive trait-specific shared pattern, which means the two methods are being used to index opposite ends of the same variance decomposition: ISC the shared end, IS-RSA the idiosyncratic end.

ISC measures how much of a response is shared, not why it is shared

The spatial distribution of ISC is not a pure readout of stimulus content: ISC tracked a visual–transmodal connectivity gradient at the whole-brain level in two independent 7T movie datasets, and this relationship was independent of movie content, leading the authors to describe movie-watching as a brain state separable from content [4]. ISC magnitude also varies substantially with the stimulus: across three datasets totalling 318 subjects and 36 movies, ISC varied between movies throughout cortex, and this variability was spatially heterogeneous, with regions showing stronger ISC also showing greater between-movie variability [8]. Because ISC is inherently stimulus-driven, effects observed for one movie may not generalize to another [8]. ISC additionally fails to capture fine-grained idiosyncrasy: spectral decomposition of movie-viewing responses showed idiosyncratic neural patterns persisting across multiple orders of magnitude of latent dimensions, and these patterns "cannot be reduced to those detected by conventional intersubject correlation measures" [15].

IS-RSA converts individual differences into the modelled signal, but only at the level of dyadic similarity

Applying IS-RSA to 174 Human Connectome Project participants, higher neuroticism was associated with greater heterogeneity in brain responses, an effect most pronounced during independently rated trait-relevant movie scenes [9]. In dyadic work, inter-brain neural similarity within mentalizing regions tracked affective similarity only in the self-perspective condition and not in the same-stimuli condition, and under a novel friend-perspective the coupling depended on social anxiety symptoms [1]. Neural similarity computed before participants met predicted proximity in a friendship network eight months later, with pre-existing neural similarity distinguishing people who grew closer from those who drifted apart [25]. Representational geometry reconstructed from naturalistic viewing also predicted inferred personality traits [13]. What unites these results is that the trait enters as a predictor of pairwise similarity, not as a difference in mean response—so the shared stimulus response is absorbed into the similarity metric rather than separated from it.

Trait-linked structure shows up most reliably as deviation from consensus

In 82 healthy subjects, lower typicality of brain activity during naturalistic viewing was the most consistent neural correlate of higher Neuroticism/Harm Avoidance [6]. Typicality behaves like a stable individual-difference measure rather than a nuisance quantity: behavioural and neural typicality were highly stable within individuals while remaining sensitive to the computations engaged by different stimuli, and neural typicality predicted objectively measured performance in motion prediction and face recognition [28]. Clinical comparisons follow the same logic: healthy–healthy dyads showed significantly higher ISC than intermittent-explosive-disorder dyads in default mode and salience regions, even after controlling for self-reported emotional responses and intention attributions [24]; individuals with schizophrenia showed less normative and more idiosyncratic responses, extending beyond the DMN into limbic and frontoparietal control networks [2]; and ISC during emotional versus neutral clips isolated reduced bilateral amygdala synchrony in patients from general visual processing [5]. The methodological consequence is that "stable individual differences" are being inferred from the absence of shared response, which makes the inference vulnerable to any factor that reduces synchrony without being a trait.

Alignment shows that much apparent idiosyncrasy is topographical, not content-related

Response hyperalignment in 12 macaques greatly improved ISC of movie responses in occipital, temporal, and prefrontal cortex and afforded accurate between-subject decoding of movie time-points despite idiosyncratic functional topographies, demonstrating a shared high-dimensional representational space [19]. Hyperalignment-derived topographies predicted individuals' own localizer topographies, recovered idiosyncratic category-selective features, outperformed anatomical alignment, and generalized across groups including developmental prosopagnosia [18]. Most decisively for the stability question, hyperalignment was used to decompose the heritability of movie-evoked activity into genetic similarity in where versus how sensory information is processed, with heritability concentrated in lower temporal frequencies and in more associative cortex and partly explained by heritable neural timescale [14]. Functional alignment and template construction also determine cross-subject decoding accuracy, with optimal transport templates performing best and showing no significant bias toward individual subjects [16]. Idiosyncrasy that survives alignment is a candidate stable individual difference; idiosyncrasy that dissolves is topography.

The separating operation is a contrast or a design, not the correlation coefficient

Group differences in stimulus-driven coupling within the DMN in schizophrenia were present for inter-subject functional connectivity but absent for within-participant functional connectivity, indicating that the effect reflects reduced stimulus-driven coordination rather than a general decline in connectivity [3]. Comparing emotional with neutral clips removed general visual processing to reveal amygdala-specific synchrony differences [5], and rating scenes for trait-relevance allowed IS-RSA effects to be localized to trait-relevant content [9]. Individual differences during movie-watching also have an intrinsic, gaze-mediated source: during low-saliency periods, when inter-individual variability was greatest, right temporo-parieto-occipital junction activity predicted visual features of upcoming gaze targets two seconds before eye movements [23]. Some inter-individual structure is furthermore task-invariant, with whole-brain inter-individual correlation patterns remarkably consistent across seven tasks and rest [7].

Conflicts and gaps

The corpus disagrees on content-independence. ISC topography tracked a cortical gradient independent of movie content [4], yet ISC and ISC–behaviour associations varied markedly between movies, with limited spatial overlap of IS-RSA effects across movies for two distinct behavioural constructs [8]. These may be claims at different levels (regional topography versus individual-difference coupling), but no retrieved paper tests that reconciliation directly.

Whether reduced synchrony is trait signal or measurement failure is unresolved. Observed standardized group differences are attenuated as a function of the square root of reliability [12], and none of the retrieved ISC/IS-RSA studies reports cross-session test–retest reliability of its individual-difference estimate, so "stable" is largely inferred from single-session designs. Stability evidence exists for typicality [28] and for prospective neural-similarity prediction [25], but not for the ISC–behaviour slope itself.

Metric dependence is unexamined in this literature. Conclusions about representational correspondence depend strongly on the chosen similarity measure, with models best under one measure often worst under another [27]; geometry-preserving metrics such as CKA and Procrustes separate structure more reliably than linear predictivity [17], and metric families can be fused to sharpen separation [26]. RSA and mean-centred CKA are largely equivalent once centring is applied [20], yet no retrieved IS-RSA study reports a multiverse across similarity metrics.

Classic RSA cannot weight individual trials or rows/columns of a similarity matrix, and multilevel trial-level RSA was both more theoretically appropriate and more sensitive to true effects [10]; the retrieved IS-RSA studies all use the classical, non-multilevel form. Finally, gaze is rarely modelled: concurrent eye-tracking databases exist [22] and fMRI-derived gaze estimates are reliable at the group level [21], but the clinical ISC studies control for self-reported affect [24] rather than for differential looking.

What would change the answer

A two-session, two-movie design in the same participants—estimating the latent neural-similarity/trait-similarity correlation with Bayesian two-task hierarchical models, which recover latent correlations more accurately than conventional correlations [11]—would directly test whether the IS-RSA component is stable or movie-specific. Extending the multi-movie design [8] to out-of-sample prediction (fit the similarity–behaviour slope on movies 1..k, predict behaviour from held-out movies) would adjudicate the disagreement with content-independent ISC topography [4]. Applying the twin/heritability decomposition [14] to the ISC and IS-RSA individual-difference components would supply an external stability criterion independent of the same-session data. Reanalyzing existing IS-RSA datasets with multilevel trial-level models [10] and a multiverse across similarity metrics [27] would establish how much of the reported trait coupling is analytic. Finally, datasets with synchronized eye-tracking [22] analysed with gaze as an explicit covariate would test whether idiosyncratic neural responses reflect individual neural processing or individual looking.

References

[1] Alicia Vallorani, Kathryn McNaughton, Marisa Lytle, Michael Hallquist, Elizabeth Redcay, Koraly Perez-Edgar. "Mentalizing from self and friend perspectives: Relations between neural similarity, affective similarity and social anxiety symptoms". psyarxiv. 2026-07-15T15:12:28.390Z. URL: https://osf.io/z5ebw_v2

[2] Yixuan Lisa Shen, Yizhou Lyu, LOURDES ESPARZA, Ioana Ciuperca, Michael Green, Amy Jimenez, Catherine Sugar, Carolyn Parkinson, ERIC REAVIS. "Naturalistic Social Processing Is Less Normative and More Heterogeneous in Schizophrenia". psyarxiv. 2026-08-08T16:25:14.201Z. URL: https://osf.io/q7y92_v1

[3] Lyu, Y., Shen, Y. L., Esparza, L. C., Reavis, E. A., Parkinson, C.. "Reduced Functional Coordination within the Default Mode Network in Schizophrenia During Naturalistic Neuroimaging". bioRxiv. 2026-08-20T00:00:00.000Z. URL: https://doi.org/10.64898/2026.08.11.744321

[4] Smith, M. E., Samara, A., Cabalo, D. G., Ngo, A., Shearer, H., Bernhardt, B. C., Vanderwal, T.. "Patterns of intersubject correlations parallel organizational gradients during naturalistic viewing". bioRxiv. 2025-10-20T00:00:00.000Z. URL: https://doi.org/10.1101/2025.10.19.683222

[5] Pallavicini, C., Costanzo, E. Y., de la Fuente, L. A., Castro, M. N., Guinjoan, S. M., Tagliazucchi, E., Villarreal, M.. "Disrupted Emotional Neural Synchrony in Schizophrenia Revealed by Intersubject Correlation of Naturalistic fMRI". bioRxiv. 2026-04-14T00:00:00.000Z. URL: https://doi.org/10.64898/2026.04.13.718247

[6] Jajcay, L., Tomecek, D., Androvicova, R., Fajnerova, I., Dechterenko, F., Rydlo, J., Tintera, J., Lukavsky, J., Horacek, J., Hlinka, J.. "Deviation from typical brain activity during naturalistic stimulation is related to personality traits". bioRxiv. 2025-11-11T00:00:00.000Z. URL: https://doi.org/10.1101/2024.04.23.586759

[7] Di, X., Jain, P., Biswal, B. B.. "Effects of Tasks on Functional Brain Connectivity Derived from Inter-Individual Correlations: Insights from Regional Homogeneity of Functional MRI Data". bioRxiv. 2025-05-05T00:00:00.000Z. URL: https://doi.org/10.1101/2024.06.02.597063

[8] Leipold, S., Ravi Rao, R., Schoffelen, J.-M., Bogels, S., Toni, I.. "Inter-subject correlations and their behavioral associations vary across movies: Implications for generalizability". bioRxiv. 2026-05-22T00:00:00.000Z. URL: https://doi.org/10.1101/2024.12.03.626542

[9] Popp, J. L., Weiss, M., Faskowitz, J., Hilger, K.. "The Neural Code of Neuroticism". bioRxiv. 2026-01-19T00:00:00.000Z. URL: https://doi.org/10.64898/2026.01.19.700296

[10] Huang, S., Howard, C. M., Bogdan, P. C., Morales-Torres, R., Slayton, M., Clarke, A., Cabeza, R., Davis, S. W.. "Trial-level Representational Similarity Analysis". bioRxiv. 2026-08-21T00:00:00.000Z. URL: https://doi.org/10.1101/2025.03.27.645646

[11] Cui Ding, Lena Jäger. "Measurement Reliability of Individual Differences in Naturalistic Reading". psyarxiv. 2026-08-17T06:44:22.409Z. URL: https://osf.io/gak3m_v1

[12] Povilas Karvelis, Andreea Diaconescu. "Clarifying the reliability paradox: poor measurement reliability attenuates group differences". psyarxiv. 2025-03-30T03:04:11.254Z. URL: https://osf.io/z4yqe_v4

[13] Kwon, D., Jolly, E., Chang, L. J., Shim, W. M.. "Neural geometry of social knowledge emerging from observed social interactions". bioRxiv. 2026-08-19T00:00:00.000Z. URL: https://doi.org/10.64898/2026.02.01.703160

[14] Gruskin, D. C., Vieira, D. J., Lee, J. K., Patel, G. H.. "Heritability of movie-evoked brain activity and connectivity". bioRxiv. 2026-04-30T00:00:00.000Z. URL: https://doi.org/10.1101/2024.09.16.612469

[15] Chihye Han, Michael F. Bonner. "High-dimensional structure underlying individual differences in naturalistic visual experience". arxiv. 2026-02-05T00:00:00.000Z. URL: https://doi.org/10.1016/j.cub.2025.12.039

[16] Barbarant, P.-L., Meyniel, F., Thirion, B.. "Functional Templates in fMRI: Building Accurate and Interpretable Group-Level Decoders". bioRxiv. 2026-06-15T00:00:00.000Z. URL: https://doi.org/10.64898/2026.05.21.726781

[17] Yiqing Bo, Ansh Soni, Sudhanshu Srivastava, Meenakshi Khosla. "Evaluating Representational Similarity Measures from the Lens of Functional Correspondence". arxiv. 2025-09-16T00:00:00.000Z. URL: https://arxiv.org/abs/2411.14633

[18] Abenes, I., Jiahui, G.. "Predicting Individualized Functional Topography in Developmental Prosopagnosia". bioRxiv. 2026-03-20T00:00:00.000Z. URL: https://doi.org/10.64898/2026.03.18.712539

[19] Zhang, Y., Wang, H., Zhu, Q., Li, X., Han, J., Jiahui, G., Shim, W. M., Kim, R., Gobbini, M. I., Heo, K., Sepe, A., Haxby, J. V., Feilong, M., Vanduffel, W.. "Hyperalignment Reveals Shared Information in Idiosyncratic Fine-Grained Movie fMRI Patterns in Macaques". bioRxiv. 2026-08-29T00:00:00.000Z. URL: https://doi.org/10.64898/2026.06.14.731832

[20] Williams, A. H.. "Equivalence between representational similarity analysis, centered kernel alignment, and canonical correlations analysis". bioRxiv. 2024-10-24T00:00:00.000Z. URL: https://doi.org/10.1101/2024.10.23.619871

[21] Gao, L., Wei, Z., Biswal, B. B., Di, X.. "fMRI-Based Prediction of Eye Gaze During Naturalistic Movie Viewing Reveals Eye-Movement-Related Brain Activity". bioRxiv. 2026-04-12T00:00:00.000Z. URL: https://doi.org/10.64898/2026.01.10.698820

[22] Levchenko, E., Chow-Wing-Bom, H., Dick, F., Tierney, A. T., Skipper, J. I.. "A neuroimaging database combining movie-watching, eye-tracking, sensorimotor mapping, and cognitive tasks". bioRxiv. 2025-10-24T00:00:00.000Z. URL: https://doi.org/10.1101/2025.09.25.678556

[23] Inoue, M., Oizumi, M., Sasai, S.. "Seeing from the inside out: Intrinsic neural dynamics explain individual differences in movie watching". bioRxiv. 2025-12-08T00:00:00.000Z. URL: https://doi.org/10.64898/2025.12.03.692223

[24] Chen, J., Keedy, S., Coccaro, E., Leong, Y. C.. "Individuals with Intermittent Explosive Disorder Exhibit Idiosyncratic Neural Responses during Social-emotional Processing". bioRxiv. 2026-03-17T00:00:00.000Z. URL: https://doi.org/10.64898/2026.03.13.711681

[25] Yixuan Lisa Shen, Ryan Hyon, Thalia Wheatley, Adam Kleinbaum, Christopher Welker, Carolyn Parkinson. "Neural Similarity Predicts Whether Strangers Become Friends". psyarxiv. 2025-03-06T18:33:33.137Z. URL: https://osf.io/vyge6_v1

[26] Jialin Wu, Shreya Saha, Yiqing Bo, Meenakshi Khosla. "Comparing and Integrating Different Notions of Representational Correspondence in Neural Systems". arxiv. 2026-02-24T00:00:00.000Z. URL: https://arxiv.org/abs/2509.21628

[27] Soni, A., Srivastava, S., Maechler, M. R., Khosla, M., Kording, K. P.. "Conclusions Drawn From Neural Network to Brain Alignment Depend Strongly on the Chosen Similarity Measure". bioRxiv. 2026-07-08T00:00:00.000Z. URL: https://doi.org/10.1101/2024.08.07.607035

[28] Zamberg-Elad, M., Har-Shalom, I., Jarbi, A., Wilf, M., Ramot, M.. "From Consensus to Individual Differences: Typicality Links Brain and Behavior Across Naturalistic Contexts". bioRxiv. 2026-10-19T00:00:00.000Z. URL: https://doi.org/10.64898/2026.08.14.744685

How this survey was made

This survey answers one question: “How do inter-subject correlation and inter-subject representational similarity analysis of naturalistic fMRI distinguish shared stimulus-driven responses from stable individual differences in brain activity?”

We keep our own collection of recent preprints and refresh it every day from bioRxiv, medRxiv, arXiv, PsyArXiv, SocArXiv and ChemRxiv. For each survey, we search that collection for recent work and keep the strongest matches: papers that use the same terms as the question, and papers on the same topic in different words.

The text was written from the titles and abstracts of those papers; the full texts were not read. Statements that draw on a particular paper are marked with a number in brackets. The reference list just above is the 28 papers whose work this text draws on, chosen from the 139 papers those searches returned; the rest of the shortlist is not printed. It is not a complete review of everything published on the question.

Treat this as a starting point and a reading list, not a peer-reviewed review. The searches behind this survey ran on October 21, 2026 and covered work from the previous two years.