Creating candidate networks from multiple synconset waves.

<p>(<b>a</b>) The three panels show the observed latencies in response to three different stimuli. Based on the above latency profiles, a reconstruction is to be made. (<b>b</b>) A wiring probability to create candidate network. One of the ways of constructing candidate networks is based on the observation that neurons with small difference in their latencies are more likely to be connected. The graph in this figure shows the connection probability of two neurons as a function of their latency difference (). This particular probability scheme always connects a neuron to another one with a higher latency as is evident from the zero probability at negative latency differences. Highest connection probability is to a neuron firing a little later(5 ms) than itself. The probability drops off exponentially on both sides of this maximum. (<b>c</b>) Three partial reconstructions of the network corresponding to the three synconsets in (a). The probability profile in (b) was used for this reconstruction. The colours of neurons show the activation latencies as shown in the key. Synapses(edges) take the colours of the pre-synaptic neurons. (<b>d</b>) The reconstructed candidate network. This reconstruction is just a merge of the three partial reconstructions in (c). If a synapse exists in any one of the three partial networks in (c), it exists in the merged network too. Notice that this reconstructed network has no edges in the bottom right of the network. This is due to the fact that none of the 3 stimulations of the original network in (a) activated this part of the network. Hence we have no information about it. (<b>e)</b> The predicted latency response to the three stimulations in the reconstructed candidate network of (d). Note that the colours now represent the predicted latencies in the new network and not the original latencies as in (a) and (c). Note that these predicted latencies also have a centre out structure as in the original network. Although the detailed structure of latencies is different, the likelihood values for the reconstructed network are of the order of the original network.