Systematic Assessment of Deep Learning-Based Predictors of Fragmentation Intensity Profiles.

In recent years, several deep learning-based methods have been proposed for predicting peptide fragment intensities. This study aims to provide a comprehensive assessment of six such methods, namely Prosit, DeepMass:Prism, pDeep3, AlphaPeptDeep, Prosit Transformer, and the method proposed by Guan et al. To this end, we evaluated the accuracy of the predicted intensity profiles for close to 1.

Rigorous treatment of pairwise and many-body electrostatic interactions among dielectric spheres at the Debye-Hückel level.

Electrostatic interactions among colloidal particles are often described using the venerable (two-particle) Derjaguin-Landau-Verwey-Overbeek (DLVO) approximation and its various modifications. However, until the recent development of a many-body theory exact at the Debye-Hückel level (Yu in Phys Rev E 102:052404, 2020), it was difficult to assess the errors of such approximations and impossible to assess the role of many-body effects. By applying the exact Debye-Hückel level theory, we quantify the errors inherent to DLVO and the additional errors associated with replacing many-particle interactions by the sum of pairwise interactions (even when the latter are calculated exactly).

Extending electrostatics of dielectric spheres to arbitrary charge distributions with applications to biosystems.

A previously developed classical model of electrostatic interactions, based on a formalism of dielectric spheres, which has been found to have surprising accuracy for S state atoms, is extended by allowing higher-order moments of the intrinsic charge distribution. Two methods to introduce the charge distribution (point moments at the center vs surface charge) are shown to be equivalent and are compared with another common model for polarizable atoms that utilizes polarizable point dipoles. Unlike the polarizable point dipole model, the polarizable spheres models do not suffer from a divergence at small separation of atoms and are easily generalized to higher multipoles.

Can dielectric spheres accurately model atomic-scale interactions?

We calculate the polarization portion of electrostatic interactions at the atomic scale using quantum mechanical methods such as density functional theories (DFT) and the coupled cluster approach, and using classical methods such as a surface charge method and a polarizable force field. The agreement among various methods is investigated. Using the coupled clusters method CCSD(T) with large basis sets as the reference, we find that for systems comprising two to six atoms and ions in S-states the classical surface charge method performs much better than commonly used DFT methods with moderate basis sets such as B3LYP/6-31G(d,p).

Using dissociation energies to predict observability of b- and y-peaks in mass spectra of short peptides. II. Results for hexapeptides with non-polar side chains.

Rationale: The hypothesis that dissociation energies can serve as a predictor of observability of b- and y-peaks is tested for seven hexapeptides. If the hypothesis holds true for large classes of peptides, one would be able to improve the scoring accuracy of peptide identification tools by excluding theoretical peaks that cannot be observed in practical product ion spectra due to various physical, chemical or thermodynamic considerations.

Methods: Product ion m/z spectra of hexapeptides AAAAAA, AAAFAA, AAAVAA, AAFAAA, AAVAAA, AAFFAA and AAVVAA have been acquired on a Finnigan LTQ XL mass spectrometer in the collision-induced dissociation (CID) activation mode on a grid of activation times 0.