High Precision Prediction of Functional Sites in Protein Structures

Abstract

We address the problem of assigning biological function to solved protein structures. Computational tools play a critical role in identifying potential active sites and informing screening decisions for further lab analysis. A critical parameter in the practical application of computational methods is the precision, or positive predictive value. Precision measures the level of confidence the user should have in a particular computed functional assignment. Low precision annotations lead to futile laboratory investigations and waste scarce research resources. In this paper we describe an advanced version of the protein function annotation system FEATURE, which achieved 99% precision and average recall of 95% across 20 representative functional sites. The system uses a Support Vector Machine classifier operating on the microenvironment of physicochemical features around an amino acid. We also compared performance of our method with state-of-the-art sequence-level annotator Pfam in terms of precision, recall and localization. To our knowledge, no other functional site annotator has been rigorously evaluated against these key criteria. The software and predictive models are incorporated into the WebFEATURE service at http://feature.stanford.edu/wf4.0-beta.

Improving the explainability of Random Forest classifier – user centered approach

“…FEATURE data i.e. the training database used for RF training, contains feature vectors at known positive (functional site) and negative (background) class labels for each protein functional model [18]. FEATURE training data is highly imbalanced e.g.…”

“…there are two to three orders of magnitude more negative (background) vs. positive (functional sites) samples. For the work in this paper we used the same 7 FEATURE models selected in experiments in [5], which are subset of models analyzed in [18], see Table 1.…”

“…Common problem with this approach is still a large number of complex rules hard to interpret by humans and lack of tradeoffs between accuracy and number of rules used. Our prior work on explainability for RF was motivated by our original joint work with Stanford Helix team on applying Support Vector Machines (SVM) [18] and RF [5] to their FEATURE data [19] where we show very good classification results measured by high recall and precision. In [5] we made first attempts to improve explainability by using RF-provided variable importance measures but did not analyze positive vs. negative classes separately and achieved very limited explainability improvements.…”

Using support vector machines to improve elemental ion identification in macromolecular crystal structures

“…They have also proven useful for identifying possible metal-binding sites from structure alone (Bordner, 2008;Buturovic et al, 2014). Here, we trained SVMs on information from the X-ray scattering and local chemical environment.…”

“…In the context of structural biology, these methods have shown success in the analysis of crystallization images (Pan et al, 2006) as well as in the prediction of binding and functional sites from both sequence (Lippi et al, 2012;Carugo, 2008) and structure (Brylinski & Skolnick, 2011;Buturovic et al, 2014), structural polymorphism (Takaya et al, 2013), the results of mutation experiments (Wei et al, 2013) and model building into electron density (Holton et al, 2000;Gopal et al, 2007). Here, we present an advance upon our previous method, in which we use support vector machines (SVMs) to classify sites as either water or one of various elemental ions.…”