Algorithmic Prediction of Binding Sites of TNFα/TNFR2 and PD-1/PD-L1

Predicting the binding sites of proteins has historically relied on the determination of protein structural data. However, the ability to utilize binding data obtained from a simple assay and computationally make the same predictions using only sequence information would be more efficient, both in time and resources. The purpose of this study was to evaluate the effectiveness of an algorithm developed to predict regions of high-binding on proteins as it applies to determining the regions of interaction between binding partners. This approach was applied to tumor necrosis factor alpha (TNFα), its receptor TNFR2, programmed cell death protein-1 (PD-1), and one of its ligand PD-L1. The algorithms applied accurately predicted the binding region between TNFα and TNFR2 in which the interacting residues are sequential on TNFα, however failed to predict discontinuous regions of binding as accurately. The interface of PD-1 and PD-L1 contained continuous residues interacting with each other, however this region was predicted to bind weaker than the regions on the external portions of the molecules. Limitations of this approach include use of a linear search window (resulting in inability to predict discontinuous binding residues), and the use of proteins with unnaturally exposed regions, in the case of PD-1 and PD-L1 (resulting in observed interactions which would not occur normally). However, this method was overall very effective in utilizing the available information to make accurate predictions. The use of the microarray to obtain binding information and a computer algorithm to analyze is a versatile tool capable of being adapted to refine accuracy.