Computational Methods in Social Neuroscience: Recent Advances and Future Directions

Introduction to Computational Social Neuroscience

• Conceptual Overview: The human brain faces significant computational challenges when navigating complex social environments, which is a factor believed to have shaped its evolution according to Dunbar (2003).
• Methodological Integration: Understanding how the brain addresses these challenges requires tools capable of capturing the multivariate, interdependent nature of social thought and behavior. This involves integrating theories from social psychology and neuroscience with methods from network science and machine learning.
• Strategic Goal: By combining computational methods for characterizing social environments (stimuli, paradigms, and social relationship webs) with methods for capturing psychological processes and neuroimaging data, researchers can gain insights into how individuals navigate social worlds.

Modeling Social Decisions and Behaviors

• Parallel Computations: According to Molapour et al. (2021), the brain must perform several parallel computations during social interactions, such as tracking others’ identities, intentions, and behaviors. It is critical to understand how the brain integrates these computations across various modalities.
• Core Computational Modeling Process:     * Researchers create a model of a social decision or learning process.     * They relate the parameters of that model to neural responses measured via functional magnetic resonance imaging (fMRI).     * Example: Park et al. (2021) used this approach to study the role of the right temporoparietal junction (rTPJ) in updating impressions of strangers versus close others.
• Reinforcement Learning (RL) Framework: Lockwood and Klein-Fl%%gge (2021) provided a primer on fitting RL models to social data, offering tutorials and code to assist in deriving new hypotheses about everyday social navigation.
• Ethology and Behavioral Ecology Approaches:     * Mobbs et al. (2018) advocate for contextualizing social decisions in scenarios individuals face in natural environments, rather than using decontextualized paradigms.     * Marginal Value Theorem (MVT): Originally proposed by Charnov (1976), MVT characterizes decisions regarding when to abandon a current location to move to a new setting when foraging for rewards. Gabay and Apps (2021) suggest conceptualizing social cognition as "foraging for social information," using MVT to quantify behavior in complex, life-like situations.

Probing Neural Representations with Multivariate Approaches

• Multivoxel Pattern Analysis (MVPA):     * Utility: MVPA provides sensitivity to information in distributed neural patterns and reveals how brain regions organize information (the rules governing similarity or distinction between stimuli).     * Representational Similarity Analysis (RSA): A specific MVPA technique discussed by Popal et al. (2019) for examining the structure of neural representations.
• Spatial Considerations: Jolly and Chang (2021) discuss the spatial scales of MVPA, emphasizing the differences between:     * Searchlight analyses.     * Region-of-interest (ROI) analyses.     * Whole-brain predictive modeling.
• Specific Model Applications:     * ACT-FAST Model: Thornton and Tamir (2021) propose this framework for how the brain represents observed action sequences to predict future actions.     * Dimensional Value Mapping: Londer%%e and Wagner (2021) used RSA to show the orbitofrontal cortex (OFC) encodes multiple value dimensions (e.g., food tastiness vs. healthiness). This suggests similar methods could probe how dimensions of "social value" organize our perceptions of partners.     * Social Perception: Brooks et al. (2021) review how MVPA identifies the transformation of social information at different processing stages.

Establishing Causal Links via Decoded Neurofeedback

• Limitations of Tradition fMRI: Usually, fMRI only establishes correlational links between brain activity and behavior.
• fMRI-Based Decoded Neurofeedback: This method establishes causality by targeting specific multivoxel response patterns.     * Mechanism: Participants are given real-time feedback on the activation likelihood of a targeted neural pattern. The elicitation of this pattern is associated with a reward.     * Awareness: Participants are unaware of what the patterns represent (e.g., they can be trained to evoke patterns associated with spiders without ever seeing a spider).     * Results: This approach has successfully reduced fear (Koizumi et al., 2017; Taschereau-Dumouchel et al., 2018) and influenced metacognition, threat reactivity, learning, and emotion perception.
• Granularity of Causality: Unlike lesion studies or Transcranial Magnetic Stimulation (TMS), which link entire regions to behavior, decoded neurofeedback links specific patterns or functional connectivity patterns (Ramot et al., 2017) to psychological outcomes.

Connectivity and the Social Brain

• Connectome-Based Fingerprinting: Finn et al. (2015) identified patterns of functional connectivity that serve as diagnostic identities for individuals. These predict:     * Fluid intelligence.     * Attentional abilities.     * Personality traits (Hsu et al., 2018).     * Psychopathology dimensions (Xia et al., 2018).     * Social relationship structures (Hyon et al., 2020b).
• Time-Varying (Dynamic) Connectivity: Calhoun et al. (2014) and Iraji et al. (2021) discuss capturing how brain states evolve over time (the ‘chronnectome’).
• Edge-Centric Representations: Focus on how the connections themselves interact (Faskowitz et al., 2020).
• Structure-Function Relationships: Tovar and Chavez (2021) demonstrated that the structural connectivity of the medial prefrontal cortex (MPFC) constrains its functional profile, with similar parcellations found using both coactivation and anatomical methods.
• Naturalistic vs. Resting State: Characterizing connectivity during movie-viewing yields more reliable estimates and better behavioral prediction than resting-state data, particularly when movies contain social content like faces and dialog (Finn and Bandettini, 2021).

Examining the Brain in Social Contexts

• Naturalistic Neuroimaging:     * Traditional tasks prioritize control but strip away context. Naturalistic studies use movies or stories where meaning unfolds over time (Sonkusare et al., 2019).     * Temporal Receptive Windows (TRW): Many social brain regions (like the Default Mode Network) are attuned to information unfolding over minutes rather than milliseconds (Hasson et al., 2008).
• Inter-Subject Correlation (ISC): Links neural time series similarity to shared subjective understanding, emotional responding, or social proximity (Parkinson et al., 2018).
• Social Interactions and Hyperscanning:     * Methods: fMRI hyperscanning (two people in separate scanners communicating), EEG, and fNIRS.     * fNIRS Advantages: Portability allows for data collection in diverse settings (e.g., classrooms, non-WEIRD samples) and permits free movement (Burns and Lieberman, 2019).
• Social Network Analysis (SNA):     * Integrates social neuroscience with real-world network data to see how people are shaped by their social ties.     * Baek et al. (2021) provide tutorials on combining neuroscience with SNA to implement key concepts in practical research.

Questions & Discussion

• What are the key trade-offs in naturalistic neuroimaging?: While researchers sacrifice experimental control because variables vary in tandem, they gain ecological validity and the ability to observe brain activity in the narrative and temporal contexts typical of everyday life.
• How does structure relate to function in the MPFC?: Evidence suggests functional coactivation patterns in the medial prefrontal cortex reflect its underlying structural connectivity, supporting the idea that anatomy constrains functional roles in social and affective phenomena.
• What is the benefit of fNIRS over traditional fMRI for social research?: fNIRS equipment is portable, allowing for interaction in more natural settings where subjects can move and express themselves freely, unlike the constrained environment of an fMRI scanner.
• How do social networks relate to neural activity?: Evidence suggests that proximity in a real-world social network (e.g., friendship) is associated with similar neural response trajectories when viewing naturalistic stimuli.