Table 1 Training datasets

No.	Topology	Time series length [samples]	Range of rates [s⁻¹]	SNR	Train size	Validation size	Test size	Generation	Noise type	Step response
1	Linear Five-State	10 M	10²–10⁵	5	10 M	180 k	180 k	simulated	experimental (patch)	experimental
2	COCOC	10 M	10²–10⁵	5	980 k	10 k	10 k	simulated	experimental (patch)	experimental
3	CCCOO	10 M	10²–10⁵	5	980 k	10 k	10 k	simulated	experimental (patch)	experimental
4	CCCOO	100 M	10²–10⁵	5	980 k	10 k	10 k	simulated	experimental (patch)	experimental
5	COCOC	10 M	10²–10⁵	2	980 k	10 k	10 k	simulated	experimental (patch)	experimental
6	COCOC	10 M	10²–10⁴ 10⁴–10⁶	5	980 k	10 k	10 k	simulated	experimental (patch)	experimental
7	COCOC	1 M	10²–10⁵	4–6	980 k	10 k	10 k	simulated	experimental (bath)	experimental
8	COCOC	1 M	10²–10⁵	4–6	980 k	10 k	10 k	simulated	experimental (bath)	4-pole Bessel
9	COCOC	1 M	10²–10⁵	4–6	980 k	10 k	10 k	simulated	lp-filtered white	experimental
10	COCOC	1 M	10²–10⁵	4–6	980 k	10 k	10 k	simulated	lp-filtered white	4-pole Bessel
11	COCOC	1 M	10²–10⁵	8–10	980 k	10 k	10 k	simulated	experimental (bath)	experimental
12	COCOC	1 M	10²–10⁵	8–10	980 k	10 k	10 k	simulated	experimental (bath)	4-pole Bessel
13	COCOC	1 M	10²–10⁵	8–10	980 k	10 k	10 k	simulated	lp-filtered white	experimental
14	COCOC	1 M	10²–10⁵	8–10	980 k	10 k	10 k	simulated	lp-filtered white	4-pole Bessel
15	COCOC	1 M	10²–10⁵	6	—	—	100	patch-clamp setup	—	—
16	COCOC	1 M	10²–10⁵	8	—	—	100	patch-clamp setup	—	—

The table lists the datasets that were used for training and evaluating the neural networks (NNs) and the parameters used for simulating data: Topology of the Markov model, length of the time series, range of the rate constants k_ij, signal-to-noise ratio (SNR), size of the training dataset, and size of the dataset used for validation and testing. Generation refers to the acquisition of time series, which is either simulated or generated on a patch-clamp setup using ideal time series as voltage commands. The last columns state which noise and step response were used for the time series simulation: experimental noise, low-pass filtered white noise, step response of a digital 4-pole Bessel filter, or experimental step response (see methods).

Search