Towards Improving Accuracy Of Protein Content Prediction For Low Homology Sequences