Unsupervised Approach for Electric Signal Separation in Gnathonemus petersii: Linking Behavior and Electrocommunication

Abstract

The transfer of information between individuals is fundamental to living systems. Therefore, communication should be studied in various species. Weakly electric fish, Gnathonemus petersii, provides a unique model organism for such investigations due to its advanced electrocommunication capabilities, using electric organ discharges (EODs). Separating EODs from multiple individuals is crucial yet challenging. To remediate it, we developed an unsupervised algorithm for EOD separation in two free-swimming individuals. Using continuous wavelet transform, t-distributed Stochastic Neighbor Embedding, and hierarchical clustering, we achieved accurate discrimination of EODs without the necessity of any training data. This approach overcomes the supervised algorithms based on previously published methods in accuracy and computational efficiency, simplifies experimental procedures, and supports animal well-being by reducing the number of required measurements. Additionally, we introduced a novel technique to map electric signals onto auditory representations, facilitating intuitive analysis of EOD sequences. These advancements lay the groundwork for future studies of EOD-based communication, highlighting the potential of Gnathonemus petersii in neuroethological, psychopharmacological, and translational research.