Prediction of order parameters based on protein NMR structure ensemble and machine learning

“For a protein consisting of 85 residues, running one time of AlphaFold structure prediction on Google’s Colaboratory takes about 3 minutes.”

“The SASA feature of the target residue is the estimated number of water molecules in contact with the residue, and calculated from protein 3D structure using the DSSP program (Kabsch and Sander 1983).”

“The most remarkable branch of machine learning is deep learning (Alzubaidi et al. 2021), which has been widely applied in areas such as image recognition, natural language processing, and more (Chao et al. 2022).”

“Random forest is a general-purpose classification and regression method proposed by L. Breiman in 2001 (Biau and Scornet 2016; Breiman 2001).”

“The model-free method is the most used NMR approach to study fast protein dynamics (Clore et al. 1990; Lipari and Szabo 1982a, 1982b), and it separates protein motion into two independent parts, namely global and internal motion.”

“Among those model-free parameters, order parameter is the most important one, which is a measure of the degree of alignment of the N-H or C-H bond vector with the principal axis of the molecular frame.”

“The model-free method is the most used NMR approach to study fast protein dynamics (Clore et al. 1990; Lipari and Szabo 1982a, 1982b), and it separates protein motion into two independent parts, namely global and internal motion.”

“…(NMR) is a powerful tool for studying the structure, dynamics, and interaction of proteins at atomic resolution (Gobl et al. 2014; Cala et al. 2014; Kleckner and Foster 2011), and can be used to investigate protein dynamics behavior across almost all time scales (Grimaldo et al. 2019; Kovermann et…”

“Machine learning is a subfield of artificial intelligence (AI) that learn patterns, relationships, and insights from data, and then use this information to make decisions or predictions without being explicitly programmed (Zhou 2021).”

“…fast motions of proteins at the picosecond to nanosec-ond timescale, known as fast dynamics, are closely related to protein conformational entropy and rearrangement, which in turn affect catalysis, ligand binding and protein allosteric effects (Jarymowycz and Stone 2006; Tzeng and Kalodimos 2011).”

“…fast motions of proteins at the picosecond to nanosec-ond timescale, known as fast dynamics, are closely related to protein conformational entropy and rearrangement, which in turn affect catalysis, ligand binding and protein allosteric effects (Jarymowycz and Stone 2006; Tzeng and Kalodimos 2011).”

“…magnetic resonance spectroscopy (NMR) is a powerful tool for studying the structure, dynamics, and interaction of proteins at atomic resolution (Gobl et al. 2014; Cala et al. 2014; Kleckner and Foster 2011), and can be used to investigate protein dynamics behavior across almost all time scales…”

“The dynamics of proteins is extremely complex, spanning several orders of magnitude in time scale, from picoseconds to seconds (Kovermann et al. 2016; Ortega et al. 2013).”

“It should be noted that, the dynamic time scale of order parameters could be extended from ps-ns to ps-ms range by measuring NH residual dipolar couplings in multiple alignments (Peti et al. 2002).”

“…spectroscopy (NMR) is a powerful tool for studying the structure, dynamics, and interaction of proteins at atomic resolution (Gobl et al. 2014; Cala et al. 2014; Kleckner and Foster 2011), and can be used to investigate protein dynamics behavior across almost all time scales (Grimaldo et al.…”

“Cilia et al proposed a linear model DynaMine which only uses protein sequence information as input to predict protein order parameters (Cilia et al. 2013).”

“These structures in the ensemble represent potential conformations that the protein can adopt, and the variations of those conformations contain rich dynamic information, as Andersen and Palmer et al pointed out 20 years ago (Andersen et al. 2002).”

“Therefore, measuring and studying protein dynamics is important for understanding the molecular mechanisms underlying protein function and related drug design (Sekhar and Kay 2019; Grimaldo et al. 2019).”

“Therefore, measuring and studying protein dynamics is important for understanding the molecular mechanisms underlying protein function and related drug design (Sekhar and Kay 2019; Grimaldo et al. 2019).”

“Berjanskii et al. introduced the RCI (random coil index) as an index to quantify the flexibility of a protein based from the secondary chemical shifts of the backbone atoms, and proposed a formula to predict order parameter from RCI (Berjanskii and Wishart 2006).”

“Zhang et al found that SASA has close relationship with residue flexibility (Zhang et al. 2009).”

“As mentioned earlier, WCN is currently the best-performing method for protein order parameter prediction (Huang et al. 2008), while cdsAF2 is the most recently proposed order parameter prediction method (Ma et al. 2023).”

“Recently, Fowler et al. also employed protein structure ensemble to assess the performance of the method they proposed for validating protein structure accuracy (Fowler et al. 2020).”

“For example, the highest Pearson correlation coefficient reported between predicted and experimental values for order parameters is 0.79, obtained by the WCN method (Ming and Bruschweiler 2006).”

“…of proteins at atomic resolution (Gobl et al. 2014; Cala et al. 2014; Kleckner and Foster 2011), and can be used to investigate protein dynamics behavior across almost all time scales (Grimaldo et al. 2019; Kovermann et al. 2016; Kleckner and Foster 2011; Gáspári and Perczel 2010; Hu and Jin 2022).”

“Similar to CM, Huang et al. proposed a weighted contact number model (WCN) to calculate 1 H-15 N order parameters only using contacted Cα atoms (Huang et al. 2008).”

“Trott et al. trained a simple neural network to predict backbone 1 H-15 N order parameters from the three-dimensional structures of proteins using a data set of 16 proteins (Trott et al. 2008).”

“…of proteins at atomic resolution (Gobl et al. 2014; Cala et al. 2014; Kleckner and Foster 2011), and can be used to investigate protein dynamics behavior across almost all time scales (Grimaldo et al. 2019; Kovermann et al. 2016; Kleckner and Foster 2011; Gáspári and Perczel 2010; Hu and Jin 2022).”

“…restrained contact map (CoRe-CMap) model for predicting protein motion from sequence, which combines residuewise predictions of protein disorder likelihood with likelihoods of inter-residue contacts obtained from contact map-based protein structure prediction methods (Vera et al. 2018).”

“For example, the highest Pearson correlation coefficient reported between predicted and experimental values for order parameters is 0.79, obtained by the WCN method (Ming and Bruschweiler 2006).”

“In a recent study, Wang et al. [110] proposed a machine learning solution that facilitates studies of fast dynamics (picosecond-nano-second timescale).”

“33–43] In particular, intriguing approaches have been reported in structure prediction [40,44] and line shape improvements [45–47] .”