Fig. 1: Schematic diagram of the high-energy atom detector (HEAD) and torsional energy descriptor (TED) modules for 3D conformation evaluation.
From: Assessing conformation validity and rationality of deep learning-generated 3D molecules
a Overview of the HEAD module. It assesses conformation validity by calculating atomic energies (step 1), establishing thresholds for anomaly detection (step 2), and classifying conformations based on energy comparisons against the threshold (step 3). The threshold is denoted as Ec. b Overview of the TED module. It evaluates conformation rationality by decomposing the structure into torsion fragments (step 1), predicting torsion energies for the target rotatable bond within each fragment (step 2), and assigning a binary label based on an energy threshold of 2 kcal/mol (step 3). Details of the torsion energy prediction neural network are provided in Methods.
