@article {Li046441, author = {Yuanning Li and Avniel Singh Ghuman}, title = {Multi-Connection Pattern Analysis: Decoding the Representational Content of Neural Communication}, elocation-id = {046441}, year = {2016}, doi = {10.1101/046441}, publisher = {Cold Spring Harbor Laboratory}, abstract = {What information is represented in distributed brain circuit interactions is unknown due to the lack of multivariate methods for decoding the representational content of interregional neural communication. Here we present Multi-Connection Pattern Analysis (MCPA), which probes the nature of the representational space contained in the multivariate functional connectivity pattern between neural populations. MCPA works by learning mappings between the activity patterns of the populations as a factor of the information being processed. These maps are used to predict the activity from one neural population based on the activity from the other population. Successful MCPA-based decoding indicates the involvement of distributed computational processing and provides a framework for probing the representational structure of the interaction. Simulations demonstrate the efficacy of MCPA in realistic circumstances. Applying MCPA to fMRI data shows that interactions between visual cortex regions are sensitive to information that distinguishes individual natural images, suggesting that image individuation occurs through interactive computation across the visual processing network. MCPA-based representational similarity analyses (RSA) are used to test hypotheses regarding information transformation between regions of the visual processing network. These results support models of error coding in interactions among regions of the network. Further RSA analyses relate the non-linear information transformation operations between layers of a computational model (HMAX) of visual processing to the information transformation between regions of the visual processing network. Thus, MCPA can be used to assess the information represented in the coupled activity of interacting neural circuits and probe the underlying principles of information transformation between regions.Significance Statement Information is represented in the brain by the coordinated activity of neurons at both the regional level and the level of large-scale, distributed networks. Multivariate methods from machine learning have advanced our understanding of the representational structure of local information coding, but the nature of distributed information representation remains unknown. Here we present a novel method that integrates multivariate connectivity analysis with machine learning classification techniques that can be used to decode the representational structure of neural interactions. This method is used to probe the representational structure of the interaction between regions of visual cortex and relate this structure to a computational model of visual processing. Thus, this work provides a framework to assess the representational content of circuit-level processing.}, URL = {https://www.biorxiv.org/content/early/2016/10/16/046441}, eprint = {https://www.biorxiv.org/content/early/2016/10/16/046441.full.pdf}, journal = {bioRxiv} }