We identify folding pathways for ubiquitin and assess its extent of transition state (TS) heterogeneity using a kinetically controlled ab initio algorithm that generates a coarse-grained description of torsional dynamics. The algorithm computes the time evolution of backbone-motion constraints, finds optimized conformations within such constraints, evaluates local solvent environments, and rescales accordingly the energetic contributions to determine the transition to the next set of torsional constraints. Native and nonnative structural features are found in the TS ensemble determined from a pool of 72 successful runs whose final folds are within 4-5Å RMSD from native. Certain nonnative features at the TS are shown to be necessary to create a large-scale context that overrides local propensities. Such misfolds undergo a subsequent rearrangement on the downhill side of the energy profile. The effects of tunable bi-histidine metal-binding sites, point mutations, negative φ-value mutations, and denaturant on kinetics are predicted.