4 Random Forest Contact Prior
The wealth of successful meta-predictors presented in section 1.2.3 highlights the importance to exploit other sources of information apart from coevolution statistics. Much information about residue interactions is typically contained in single position features that can be predicted from local sequence profiles, such as secondary structure, solvent accessibility or contact number, and in pairwise features such as the contact prediction scores for residue pairs \((i,j)\) from a simple local statistical methods as presented in section 1.2.1.
For example, predictions of secondary structure elements and solvent accessibility are used by almost all modern machine learning predictors, such as MetaPsicov [84], NeBCon [87], EPSILON-CP [86], PconsC3 [82]. Other frequently used sequence derived features include pairwise contact potentials, sequence separation and conservation measures such as column entropy [84,87,214].
In the following sections I present a random forest classifier that uses sequence derived features to distinguish contacts from non-contacts. Methods section 4.6.1 lists all features used to train the classifier including the aforementioned standard features as well as some novel features.
The probabilistic predictions of the random forest model can be introduced directly as prior information into the Bayesian statistical model that will be presented in the next section 5 to improve the overall prediction accuracy in terms of posterior probabilities. Furthermore, contact scores from coevolution methods can be added as additional feature to the random forest model in order to elucidate how much the combined information improves prediction accuracy over the single methods.
References
84. Jones, D.T., Singh, T., Kosciolek, T., and Tetchner, S. (2015). MetaPSICOV: combining coevolution methods for accurate prediction of contacts and long range hydrogen bonding in proteins. Bioinformatics 31, 999–1006.
87. He, B., Mortuza, S.M., Wang, Y., Shen, H.-B., and Zhang, Y. (2017). NeBcon: Protein contact map prediction using neural network training coupled with naïve Bayes classifiers. Bioinformatics., doi: 10.1093/bioinformatics/btx164.
86. Stahl, K., Schneider, M., and Brock, O. (2017). EPSILON-CP: using deep learning to combine information from multiple sources for protein contact prediction. BMC Bioinformatics 18, 303., doi: 10.1186/s12859-017-1713-x.
82. Skwark, M.J., Michel, M., Menendez Hurtado, D., Ekeberg, M., and Elofsson, A. (2016). Accurate contact predictions for thousands of protein families using PconsC3. bioRxiv.
214. Ma, J., Wang, S., Wang, Z., and Xu, J. (2015). Protein contact prediction by integrating joint evolutionary coupling analysis and supervised learning. Bioinformatics, btv472.