Table 5. Likely strengths and weaknesses of DeepGANnel versus traditional synthesis methods.

	Traditional Method	Traditional Method	DeepGANnel
A priori assumptions	Stochastic simulation	Everything must be estimated or assumed; channel size, rate constants, open channel noise, thermal noise levels, artefact frequency etc.	None required.
Authenticity	Stochastic simulation	Depends entirely on the accuracy of a priori-assumptions. Authenticity difficult.	Highly authentic.
Speed	Stochastic simulation	Moderate speed.	Slow to train, fast to simulate thereafter.
GPU needed	Stochastic simulation	Typically, these are not used. Future stochastic models may use them.	Realistically these are needed for training, although not for simulation itself.
	Vanilla GAN	yes	-
Markov Model	Stochastic simulation	Could include Markovian model structure.	May include Markovian structure, but this is not guaranteed.
Need for seed data	Stochastic simulation	No. The data can be completely imaginary.	Yes
Fully Labelled data	Stochastic simulation	Yes	-
	Vanilla GAN	Cannot provide labels in parallel to raw data.	Yes

This table summarises the pros and cons of DeepGANnel discussed and justified in the text. By definition such comparisons can only be subjective because Traditional Methods vary (entirely dependent on a priori assumptions, that could be simple or complex), as do computing platforms.