bookmark SPIDER / Sequence-based Prediction of Local and Nonlocal Structural Features for Proteins Predicts different sets of structural protein properties. SPIDER is an iterative deep-learning neutral network. It obtains secondary structure, torsion angles, Cα−atom based angles and dihedral angles, and solvent accessible surface area. It utilises both local and nonlocal structural information in iterations. At each iteration, SPIDER employs a deep-learning neural network to predict a structural property based on structural properties predicted in the previous iteration.