Table 8.

Model size comparison.

Variables	Median MASE
	1 GRU	5 GRU	10 GRU	20 GRU	50 GRU
CIT. BIG PHARMA	0.46	0.42	0.42	0.41	0.42
CIT. MED. PHARMA	0.51	0.50	0.47	0.48	0.48
CIT. REVIEWS	0.38	0.30	0.25	0.24	0.26
CIT. TRIALS	0.45	0.45	0.43	0.43	0.43
CITATIONS	0.39	0.26	0.21	0.23	0.25
PUB. BIG PHARMA	0.66	0.58	0.56	0.54	0.55
PUB. MED. PHARMA	0.76	0.63	0.56	0.55	0.55
PUBLICATIONS	0.48	0.33	0.26	0.26	0.26
REVIEWS	0.53	0.52	0.50	0.46	0.50

Global performance of the RNN model. Mean accuracy scaled error (MASE) is an unbiased method to compare time-series prediction models by comparing how much each model out-performs a naive model that repeats the last value. Values below 0 indicate that the method performs better than the naive model. Data for 13,380 human genes.