Fig. 1: Valence-specific model of the mushroom body.
From: Learning with reinforcement prediction errors in a model of the Drosophila mushroom body
a Schematic of several neuropils that comprise the brain of Drosophila melanogaster. The green region highlights the MB in the right hemisphere. Labels: MB mushroom body, AL antennal lobe, SOG suboesophogeal ganglion, ME medulla. b Outlines of the multiple compartments that tile the lobes of the MB, colour-coded by a broad classification of cell function. Blue: approach MBONs (mushroom body output neurons); orange: avoidance MBONs; purple: aversive DANs (dopamine neurons); green: appetitive DANs; black: KCs. Inset: schematic of the three MB lobes and their compartmentalisation. Top: \({\alpha }^{\prime}\)/\({\beta }^{\prime}\) lobes; middle: α/β lobes; bottom: γ lobe. MBON functions (approach or avoidance) are as determined in15. Pale colours in the inset correspond to MBONs that exhibit a non-significant bias on behaviour in15. c Schematic of the VS model. Units are colour-coded according to the cell types in (b). KCs connect to MBONs through plastic synapses, and connect to DANs through fixed synapses. Labels: M+—approach MBON; M−, avoidance MBON; D−, aversive DAN, D+, appetitive DAN, K, Kenyon cells, R−, negative reinforcement, R+, positive reinforcement. Lines with arrows: excitatory synapse. Lines with filled circles: synapse releasing dopamine. Downward white arrows: dopamine enhances synaptic long term depression (LTD). d Schematic of a single trial for the experimental paradigm in which the model is examined. Panel 1: the model is exposed to some number of cues that are evaluated to yield a cue-specific RP. Panel 2: using the relative RPs, the model makes a decision over which cue to choose. Panel 3: the model receives a reinforcement signal that is associated with the chosen cue, and its RPs for that cue are updated.
