I was playing with Simulate data (beta) and got some questions while making some artificial particles (and I’m new to simulating em data). Among different parameter taps, I was wondering what number is the best for signal to noise ratio that can best describe the typical experimental data (The defualt setting was 0.5). Also, how does cryosparc make and apply the noise to the data? (for example, does cryosparc first make 2d projection then perform convolution of CTF followed by applying noise or snr ratio?)
The simulate particles job works by taking slices of a 3d volume in Fourier space, applying a ctf, and then transforming back to real space, yielding a CTF corrupted projection. The noise is then applied by adding a normal distribution with standard deviation equal to the SNR times the average of all the pixels the particle stack.
Unfortunately this procedure means that the absolute value of the SNR parameter doesn’t have very much meaning. The more blank space you have around the volume, the closer the average of the particle stack to zero, and hence the lower the absolute amount amplitude of noise you will get. Further, the appearance of the noise will depend heavily on the ctf parameters you use— lower defocuses will look much worse with the same amount of noise compared with higher defocuses. The phase can also affect this.
All this to say, you will have to play with the parameters until the particles appear correct :).