We propose an improved transfer entropy approach called the dynamic version of the force constant fitted Gaussian network model based on molecular dynamics ensemble (dfcfGNMMD) to explore the allosteric mechanism of human mitochondrial phenylalanyl-tRNA synthetase (hmPheRS), one of the aminoacyl-tRNA synthetases that play a crucial role in translation of the genetic code. The dfcfGNMMD method can provide reliable estimates of the transfer entropy and give new insights into the role of the anticodon binding domain in driving the catalytic domain in aminoacylation activity and into the effects of tRNA binding and residue mutation on the enzyme activity, revealing the causal mechanism of the allosteric communication in hmPheRS. In addition, we incorporate the residue dynamic and co-evolutionary information to further investigate the key residues in hmPheRS allostery. This study sheds light on the mechanisms of hmPheRS allostery and can provide important information for related drug design.