A Multisite Study of a Breast Density Deep Learning Model for Full-Field Digital Mammography and Synthetic Mammography.

PURPOSE: To develop a Breast Imaging Reporting and Data System (BI-RADS) breast density deep learning (DL) model in a multisite setting for synthetic two-dimensional mammographic (SM) images derived from digital breast tomosynthesis examinations by using full-field digital mammographic (FFDM) images and limited SM data. MATERIALS AND METHODS: A DL model was trained to predict BI-RADS breast density by using FFDM images acquired from 2008 to 2017 (site 1: 57 492 patients, 187 627 examinations, 750 752 images) for this retrospective study. The FFDM model was evaluated by using SM datasets from two institutions (site 1: 3842 patients, 3866 examinations, 14 472 images, acquired from 2016 to 2017; site 2: 7557 patients, 16 283 examinations, 63 973 images, 2015 to 2019). Each of the three datasets were then split into training, validation, and test. Adaptation methods were investigated to improve performance on the SM datasets, and the effect of dataset size on each adaptation method was considered. Statistical significance was assessed by using CIs, which were estimated by bootstrapping. RESULTS: Without adaptation, the model demonstrated substantial agreement with the original reporting radiologists for all three datasets (site 1 FFDM: linearly weighted Cohen κ [κw] = 0.75 [95% CI: 0.74, 0.76]; site 1 SM: κw = 0.71 [95% CI: 0.64, 0.78]; site 2 SM: κw = 0.72 [95% CI: 0.70, 0.75]). With adaptation, performance improved for site 2 (site 1: κw = 0.72 [95% CI: 0.66, 0.79], 0.71 vs 0.72, P = .80; site 2: κw = 0.79 [95% CI: 0.76, 0.81], 0.72 vs 0.79, P < .001) by using only 500 SM images from that site. CONCLUSION: A BI-RADS breast density DL model demonstrated strong performance on FFDM and SM images from two institutions without training on SM images and improved by using few SM images.Supplemental material is available for this article.Published under a CC BY 4.0 license.

Effect of proline mutations on the monomer conformations of amylin.

The formation of human islet amyloid polypeptide (hIAPP) is implicated in the loss of pancreatic ß-cells in type II diabetes. Rat amylin, which differs from human amylin at six residues, does not lead to formation of amyloid fibrils. Pramlintide is a synthetic analog of human amylin that shares three proline substitutions with rat amylin. Pramlintide has a much smaller propensity to form amyloid aggregates and has been widely prescribed in amylin replacement treatment. It is known that the three prolines attenuate ß-sheet formation. However, the detailed effects of these proline substitutions on full-length hIAPP remain poorly understood. In this work, we use molecular simulations and bias-exchange metadynamics to investigate the effect of proline substitutions on the conformation of the hIAPP monomer. Our results demonstrate that hIAPP can adopt various ß-sheet conformations, some of which have been reported in experiments. The proline substitutions perturb the formation of long ß-sheets and reduce their stability. More importantly, we find that all three proline substitutions of pramlintide are required to inhibit ß conformations and stabilize the α-helical conformation. Fewer substitutions do not have a significant inhibiting effect.

α-helix to ß-hairpin transition of human amylin monomer.

The human islet amylin polypeptide is produced along with insulin by pancreatic islets. Under some circumstances, amylin can aggregate to form amyloid fibrils, whose presence in pancreatic cells is a common pathological feature of Type II diabetes. A growing body of evidence indicates that small, early stage aggregates of amylin are cytotoxic. A better understanding of the early stages of the amylin aggregation process and, in particular, of the nucleation events leading to fibril growth could help identify therapeutic strategies. Recent studies have shown that, in dilute solution, human amylin can adopt an α-helical conformation, a ß-hairpin conformation, or an unstructured coil conformation. While such states have comparable free energies, the ß-hairpin state exhibits a large propensity towards aggregation. In this work, we present a detailed computational analysis of the folding pathways that arise between the various conformational states of human amylin in water. A free energy surface for amylin in explicit water is first constructed by resorting to advanced sampling techniques. Extensive transition path sampling simulations are then employed to identify the preferred folding mechanisms between distinct minima on that surface. Our results reveal that the α-helical conformer of amylin undergoes a transformation into the ß-hairpin monomer through one of two mechanisms. In the first, misfolding begins through formation of specific contacts near the turn region, and proceeds via a zipping mechanism. In the second, misfolding occurs through an unstructured coil intermediate. The transition states for these processes are identified. Taken together, the findings presented in this work suggest that the inter-conversion of amylin between an α-helix and a ß-hairpin is an activated process and could constitute the nucleation event for fibril growth.

Ultrastable glasses from in silico vapour deposition.

Glasses are generally prepared by cooling from the liquid phase, and their properties depend on their thermal history. Recent experiments indicate that glasses prepared by vapour deposition onto a substrate can exhibit remarkable stability, and might correspond to equilibrium states that could hitherto be reached only by glasses aged for thousands of years. Here we create ultrastable glasses by means of a computer-simulation process that mimics physical vapour deposition. These stable glasses have, far below the conventional glass-transition temperature, the properties expected for the equilibrium supercooled liquid state, and optimal stability is attained when deposition occurs at the Kauzmann temperature. We also show that the glasses' extraordinary stability is associated with distinct structural motifs, in particular the abundance of regular Voronoi polyhedra and the relative lack of irregular polyhedra.

Density of states-based molecular simulations.

One of the central problems in statistical mechanics is that of finding the density of states of a system. Knowledge of the density of states of a system is equivalent to knowledge of its fundamental equation, from which all thermodynamic quantities can be obtained. Over the past several years molecular simulations have made considerable strides in their ability to determine the density of states of complex fluids and materials. In this review we discuss some of the more promising approaches proposed in the recent literature along with their advantages and limitations.

Two-dimensional infrared spectroscopy reveals the complex behaviour of an amyloid fibril inhibitor.

Amyloid formation has been implicated in the pathology of over 20 human diseases, but the rational design of amyloid inhibitors is hampered by a lack of structural information about amyloid-inhibitor complexes. We use isotope labelling and two-dimensional infrared spectroscopy to obtain a residue-specific structure for the complex of human amylin (the peptide responsible for islet amyloid formation in type 2 diabetes) with a known inhibitor (rat amylin). Based on its sequence, rat amylin should block formation of the C-terminal ß-sheet, but at 8 h after mixing, rat amylin blocks the N-terminal ß-sheet instead. At 24 h after mixing, rat amylin blocks neither ß-sheet and forms its own ß-sheet, most probably on the outside of the human fibrils. This is striking, because rat amylin is natively disordered and not previously known to form amyloid ß-sheets. The results show that even seemingly intuitive inhibitors may function by unforeseen and complex structural processes.

Efficient Free Energy Calculation of Biomolecules from Diffusion-Biased Molecular Dynamics.

Recently proposed metadynamics techniques offer an effective means for improving sampling in simulations of complex systems, including polymers and biological macromolecules. One of the drawbacks of such methods has been the absence of well-defined or effective convergence criteria. A solution to this problem is considered here in which an optimal ensemble is introduced to minimize the travel time across the entire order parameter range of interest. The usefulness of the proposed approach is illustrated in the context of two systems consisting of biological molecules dissolved in water. The results presented in this work indicate that the proposed method is considerably faster than other existing algorithms for the study of these systems, and that the corresponding free energy that emerges from the simulations converges to the exact result.

2DIR spectroscopy of human amylin fibrils reflects stable ß-sheet structure.

The aggregation of human amylin to form amyloid contributes to islet ß-cell dysfunction in type 2 diabetes. Studies of amyloid formation have been hindered by the low structural resolution or relatively modest time resolution of standard methods. Two-dimensional infrared (2DIR) spectroscopy, with its sensitivity to protein secondary structures and its intrinsic fast time resolution, is capable of capturing structural changes during the aggregation process. Moreover, isotope labeling enables the measurement of residue-specific information. The diagonal line widths of 2DIR spectra contain information about dynamics and structural heterogeneity of the system. We illustrate the power of a combined atomistic molecular dynamics simulation and theoretical and experimental 2DIR approach by analyzing the variation in diagonal line widths of individual amide I modes in a series of labeled samples of amylin amyloid fibrils. The theoretical and experimental 2DIR line widths suggest a "W" pattern, as a function of residue number. We show that large line widths result from substantial structural disorder and that this pattern is indicative of the stable secondary structure of the two ß-sheet regions. This work provides a protocol for bridging MD simulation and 2DIR experiments for future aggregation studies.

A molecular view of vapor deposited glasses.

Recently, novel organic glassy materials that exhibit remarkable stability have been prepared by vapor deposition. The thermophysical properties of these new "stable" glasses are equivalent to those that common glasses would exhibit after aging over periods lasting thousands of years. The origin of such enhanced stability has been elusive; in the absence of detailed models, past studies have discussed the formation of new polyamorphs or that of nanocrystals to explain the observed behavior. In this work, an atomistic molecular model of trehalose, a disaccharide of glucose, is used to examine the properties of vapor-deposited stable glasses. Consistent with experiment, the model predicts the formation of stable glasses having a higher density, a lower enthalpy, and higher onset temperatures than those of the corresponding "ordinary" glass formed by quenching the bulk liquid. Simulations reveal that newly formed layers of the growing vapor-deposited film exhibit greater mobility than the remainder of the material, thereby enabling a reorganization of the film as it is grown. They also reveal that "stable" glasses exhibit a distinct layered structure in the direction normal to the substrate that is responsible for their unusual properties.

Stable and metastable states of human amylin in solution.

Patients with type II diabetes exhibit fibrillar deposits of human amylin protein in the pancreas. It has been proposed that amylin oligomers arising along the aggregation or fibril-formation pathways are important in the genesis of the disease. In a step toward understanding these aggregation pathways, in this work we report the conformational preferences of human amylin monomer in solution using molecular simulations and infrared experiments. In particular, we identify a stable conformer that could play a key role in aggregation. We find that amylin adopts three stable conformations: one with an α-helical segment comprising residues 9-17 and a short antiparallel ß-sheet comprising residues 24-28 and 31-35; one with an extended antiparallel ß-hairpin with the turn region comprising residues 20-23; and one with no particular structure. Using detailed calculations, we determine the relative stability of these various conformations, finding that the ß-hairpin conformation is the most stable, followed by the α-helical conformation, and then the unstructured coil. To test our predicted structure, we calculate its infrared spectrum in the amide I stretch regime, which is sensitive to secondary structure through vibrational couplings and linewidths, and compare it to experiment. We find that theoretically predicted spectra are in good agreement with the experimental line shapes presented herein. The implications of the monomer secondary structures on its aggregation pathway and on its interaction with cell membranes are discussed.