Circular dichroism (CD) is a rapid molecular spectroscopy technique that is widely used to characterize the secondary structure of proteins. It requires only small amounts of protein and is sensitive to conformational changes resulting from varying experimental conditions. The interpretation of CD results can be significantly enhanced by correlating them with reliable spectral predictions derived from the corresponding three-dimensional structures. Such structures can be obtained from various sources, including the protein data bank, AlphaFold, or molecular dynamics simulations. This poster discusses an upgraded version of the knowledge-based circular dichroism (KCD) prediction method, referred to as KCD-AI. The core algorithm is based on the classical theory of optical activity, utilizing a series of optimized atomic polarizabilities derived from experimental data analysis. A recent implementation of a neural network allows us to determine a set of weights that produce a suitable combination of polarizabilities from a base set. This combination is used to calculate the CD spectrum of a protein under study, leading to a more accurate spectral prediction. The KCD-AI method can be employed to evaluate a protein’s secondary structure and its conformational changes under various environmental conditions, such as changes in temperature, pH, or exposure to denaturing agents. Additionally, KCD-AI can assist in assessing the reconstruction of missing strings of residues in protein crystal structures.
Carbajal-Tinoco et al. (Sun,) studied this question.