RFCRYS: sequence-based protein crystallization propensity prediction by means of random forest

Production of high-quality diffracting crystals is a critical step in determining the 3D structure of a protein by X-ray crystallography. Only 2%-10% of crystallization projects result in high-resolution protein structures. Previously, several computational methods for prediction of protein crystallizability were developed. In this work, we introduce RFCRYS, a Random Forest based method to predict crystallizability of proteins. RFCRYS utilizes mono-, di-, and tri-peptides amino acid compositions, frequencies of amino acids in different physicochemical groups, isoelectric point, molecular weight, and length of protein sequences, from the primary sequences to predict crystallizabillity by using two different databases. RFCRYS was compared with previous methods and the results obtained show that our proposed method using this set of features outperforms existing predictors with higher accuracy, MCC, and Specificity. Especially, our method is characterized by high Specificity of 0.95, which means RFCRYS rarely mispredicts a protein chain to be crystallizable which consequently would be useful for saving time and resources. In conclusion RFCRYS provides accurate crystallizability prediction for a protein chain that can be applied to support crystallization projects getting higher success rate towards obtaining diffraction-quality crystals.