Extended Data Fig. 6: Pseudo-population decoding.

(a) Odour decoding accuracy for a simultaneously recorded pPCx population (example session from Fig. 2a–d) using a wide range of time windows aligned to odour onset time (left), or first respiration after odour onset (right). The black and red dots indicate time windows used for black and red lines in (c), respectively. (b) Location decoding accuracy aligned to initiation port poke-in time across a wide range of time windows. (c) Pseudo-population decoding of odour identity with different time windows and regularization. Red and purple lines use L2 regularization, while black and grey use L1 regularization (shown as black and grey lines in Fig. 2g). By increasing the sparsity of the L1-decoder (scanning the ‘cost’ parameter over the range 2^[−7:8]) and plotting decoding accuracy as a function of the number of contributing neurons (# of neurons with non-zero weights in the decoder), we can minimize the contribution of uninformative neurons. Here, the x-axis indicates the number of neurons being used by the decoder (i.e. neurons with non-zero kernel weights), which was controlled by changing the L1 penalty. When this penalty is large, the decoder selectively uses only the most informative neurons, leading to a much steeper rise than seen for the L2 regularization pseudo-population curves, which sample neurons randomly. Using this sparse L1-decoding approach, it is clear that odour identity can accurately be decoded from a small population of pPCx neurons (90% decoding accuracy for ~150 neurons). Dotted line indicates 90% accuracy. Chance level is 25%. (d) Same analysis as in (c) but for port locations. Note that while it conveys that relatively few neurons are needed to encode odour information, the steepness of the grey curve is sensitive to the number of recorded neurons (since a larger pool is more likely to contain an informative neuron). (e, f) Yellow lines show pseudo-population decoding using neurons identified by auROC (see Methods) as responsive to either odour or location. Black and grey lines are the same as those plotted in (c, d), and Fig. 2g, reproduced here for ease of comparison.