Challenge D
Goal
Predict spike timing (with 2 ms precision) and dendritic voltage trajectories in response to current injection at the soma and up to two dendritic locations.
How to Participate
We provide a set of data for training and a set of data for evaluating the performance of your model. The training set contains the stimulation and the voltage recordings. The test set consists of a similar stimulation protocol but we keep the voltage recordings to later evaluate the performances.
Experimental Methods
For details concerning the experimental preparation, please see Larkum et al. (2004) and Larkum et al. (1999). As in challenge C, the data consists partly of noisy current injection with different means and standard deviations (Larkum et al. 2004). In this challenge, the training and test sets also contain a special stimulation protocol used to study BAC firing (Larkum et al. 1999). The injection consists of a square current pulse in the soma and an alpha-function current injected in the distal dendrite with different delays.
Evaluation Methods
In this challenge we test for two objectives which have to be satisfied simultaneously by the same model. The spike timing is evaluated with the same technique as for challenge C:
where ni corresponds to the ith recorded spike train and mi to the associated model data. Spike time is defined by the time of the peak of the somatic action potential waveform. Here, in the test set we have Nstim = 5 different stimulus conditions (experiments no. 11, 12, 14, 15, and 19). The eight spike trains to be predicted for experiment no. 19 will count as one spike train for this purpose, since they contain only a very small number of spikes (160 for experiment no. 19). Probabilistic models can submit up to 20 spike trains for each stimulus regime. The resulting performance is the average of D1 over all spike trains for all stimulus conditions. D1 is rounded to the tenth of a percent for comparison.
The second objective consists of the prediction of the proximal dendritic membrane potential to a precision of 2 mV. We use a measure related to the fraction of time the predicted voltage Vm is close to the measured voltage Vn:
where T is the total time of the stimulation, 70 ms. Nstim is the number of stimulation regimes used for testing, Nstim = 8 (all of them are found in experiment no. 19). D2 is evaluated with the proximal dendritic membrane potential (pipette D2). Again the resulting numbers are rounded to the first decimal place for comparison. The best submission is the submission that is greater or equal to the other submissions in both D1 and D2.
Participate
Download the training and test sets ChallengeD.zip You should find inside the .zip archive separate files containing the current injected, and the membrane potential recorded. The files are easily accessible, for instance you can load the membrane potential recording in MATLAB with ‘load t10D1.txt -ascii’. The membrane potential measurements are in mV, the current injections in pA and the sampling frequency is 10 kHz. Please refer to README.txt for further details.
The submission should consist of 1) 1 to 20 vectors (numbers only, ASCII) of spike times in units of 0.1 ms for each test stimulation condition, 2) your prediction of the voltage recordings in the proximal dendrites in experiment no. 19 (1 to 20 ASCII files in which the zeroed columns in t19D2.txt are replaced by the predicted membrane potential). The files should be sent to in a .tar.gz or .zip archive to : arnd dot roth at ucl.ac.uk in an email having the title ‘Challenge D submission’. We will analyze the submissions as quickly as possible and display the results with the label ‘Anonymous #xx’ until you allow us to display some details on the model used.
Submit to Arnd Roth
References
Larkum, M. E., Zhu, J. J., Sakmann, B. A new cellular mechanism for coupling inputs arriving at different cortical layers, Nature, (1999) 398: 338-341.
Larkum, M. E., Senn, W., Lüscher, H.-R. Top-Down Dendritic Input Increases the Gain of Layer 5 Pyramidal Neurons. Cerbral Cortex, (2004) 14: 1059-1070.
