Sequence-specific bias correction for RNA-seq data using recurrent neural networks

BMC Genomics

Table 2 Correlation results on Sample A, replicate 1

Method	Pearson	Spearman	r ²
Raw normalized counts	0.8644	0.8708	0.7471
seqBias	0.8658	0.8751	0.7496
GRU(4-10-4)	0.8674	0.8749	0.7523
LSTM(4-10-4)	0.8661	0.8736	0.7460
GRU(4-20-4)	0.8661	0.8729	0.7500
LSTM(4-20-4)	0.8669	0.8744	0.7515

In total, 922 genes are evaluated. RNN(a-b-c) represents an RNN model with model structure defined according to a, b, and c. The values a and c represent the node number of the input and output layer, respectively. The value b is the number of hidden units. The boldface numbers indicate the best performance in all comparisons

ISSN: 1471-2164