A simulation study of information transmission by multi-unit microelectrode recordings

Abstract

To analyse the information content of multi-unit signals, cases of two and three superimposed neural responses to a stimulus were simulated. The multi-unit Shannon information rates were compared with those of the constituent single-unit spike trains and also with those of corresponding labelled line signals.

The mutual information in the multi-unit response was found to depend on the degree of similarity in the attribute-specific information of the single-unit spike trains. As long as the units were encoding similar attribute-specific information, the multi-unit signal had greater information than any of the individual units. If the units' responses were confounded by different amounts, the multi-unit response could not recover the maximal single-unit information. Labelled line and pooled responses had similar mutual information when the information in the constituent units was similar. However, the normalized gain in information of a labelled line response over a pooled response increased from 27% to 84%, relative to the maximum single-unit information rate, as the difference in constituent single-unit information increased to 100%.

Information in single-unit responses is not completely lost when multiple units are superimposed. There are cases in which the union of single-unit spike trains can fully conserve single-unit information and even reduce the effect of confounding information present in the individual responses. More information is available in pairs of responses when the identities of the units are maintained. Summed responses can be used at low informational cost in redundantly encoding clusters of neurons. Higher information losses are incurred as the encoding becomes more independent.