Optimal dimensionality selection for independent component analysis of transcriptomic data