Background: Disulfide bonds constitute one of the most important cross-linkages in proteins and significantly influence protein structure and function. At the state-of-the-art, various methodological frameworks have been proposed for identification of disulfide bonds. These include among others, mass spectrometry-based methods, sequence-based predictive approaches, as well as techniques like crystallography and NMR. Each of these frameworks has its advantages and disadvantages in terms of pre-requisites for applicability, throughput, and accuracy. Furthermore, the results from different methods may concur or conflict in parts.
Results: In this paper, we propose a novel and theoretically rigorous framework for disulfide bond determination based on information fusion from different methods using an extended formulation of Dempster-Shafer theory. A key advantage of our approach is that it can automatically deal with concurring as well as conflicting evidence in a data-driven manner. Using the proposed framework, we have developed a method for disulfide bond determination that combines results from sequence-based prediction and mass spectrometric inference. This method leads to more accurate disulfide bond determination than any of the constituent methods taken individually. Furthermore, experiments indicate that the method improves the accuracy of bond identification as compared to leading extant methods at the state-of-the-art. Finally, the proposed framework is extensible in that results from any number of approaches can be incorporated. Results obtained using this framework can especially be useful in cases where the complexity of the bonding patterns coupled with specificities of the fragmentation pattern or limitations of computational models impair any single method to perform consistently across a diverse set of molecules.