Understanding Dynamics of Nonlinear Representation Learning and Its Application.

Representations of the world environment play a crucial role in artificial intelligence. It is often inefficient to conduct reasoning and inference directly in the space of raw sensory representations, such as pixel values of images. Representation learning allows us to automatically discover suitable representations from raw sensory data. For example, given raw sensory data, a deep neural network learns nonlinear representations at its hidden layers, which are subsequently used for classification (or regression) at its output layer. This happens implicitly during training through minimizing a supervised or unsupervised loss. In this letter, we study the dynamics of such implicit nonlinear representation learning. We identify a pair of a new assumption and a novel condition, called the on-model structure assumption and the data architecture alignment condition. Under the on-model structure assumption, the data architecture alignment condition is shown to be sufficient for the global convergence and necessary for global optimality. Moreover, our theory explains how and when increasing network size does and does not improve the training behaviors in the practical regime. Our results provide practical guidance for designing a model structure; for example, the on-model structure assumption can be used as a justification for using a particular model structure instead of others. As an application, we then derive a new training framework, which satisfies the data architecture alignment condition without assuming it by automatically modifying any given training algorithm dependent on data and architecture. Given a standard training algorithm, the framework running its modified version is empirically shown to maintain competitive (practical) test performances while providing global convergence guarantees for deep residual neural networks with convolutions, skip connections, and batch normalization with standard benchmark data sets, including MNIST, CIFAR-10, CIFAR-100, Semeion, KMNIST, and SVHN.

Perturbation effect of single polar group substitution on the Self-Association of amphiphilic peptide helices.

As an important attempt towards creating hierarchical structures more like nature, the peptide is employed as a building block to build supramolecular architectures. An emerging question is whether the molecular mechanism of self-assembly obtained from the small molecule system, e.g., the driving forces of assembly are conventionally regarded as pairwise-additive, can be manifested in the self-association of biologically relevant amphiphilic peptides. A peptide, KRT-R, was derived from the 120-144 segment of keratin 14. The single cation-to-cation substitution with KRT-R at the site of 125 from arginine (R) to either lysine (K) or histidine (H) results in the peptide helices, KRT-K and KRT-H, sharing 96% sequence identity. These KRT-derived peptides possess similarities in the folding structures but exhibit divergent self-assembled structures. KRT-R and KRT-K self-assemble into sheets and fibrils, respectively. Whereas KRT-H associates into heterogeneous structures, including sheets, particles, and branched networks. The intrinsic tyrosine fluorescence spectroscopy measurements with the KRT-derived peptides within a temperature range of 25 °C to 95 °C reveal that the heating-triggered structural transitions of KRT-derived peptides are divergent. The alternation of single cationic residue changes the thermodynamic signature of peptide assemblies upon heating. A chemical denaturation experiment with KRT-derived peptides indicates that the intermolecular interactions that govern the supramolecular architectures formed by peptides are distinct. Overall, our work demonstrates the contribution of the interplay among various noncovalent interactions to supramolecular assembly.

Trends in the biological functions and medical applications of extracellular vesicles and analogues.

Natural extracellular vesicles (EVs) play important roles in many life processes such as in the intermolecular transfer of substances and genetic information exchanges. Investigating the origins and working mechanisms of natural EVs may provide an understanding of life activities, especially regarding the occurrence and development of diseases. Additionally, due to their vesicular structure, EVs (in small molecules, nucleic acids, proteins, etc.) could act as efficient drug-delivery carriers. Herein, we describe the sources and biological functions of various EVs, summarize the roles of EVs in disease diagnosis and treatment, and review the application of EVs as drug-delivery carriers. We also assess the challenges and perspectives of EVs in biomedical applications.

Opposite Regulatory Effects of Immobilized Cations on the Folding Vs. Assembly of Melittin.

Ions are crucial in modulating the protein structure. For the free ions in bulk solution, ammonium is kosmotropic (structure forming) and guanidinium is chaotropic (structure breaking) to the protein structure within the Hofmeister series. However, the effect of immobilized ions on a protein surface is less explored. Herein, we explored the influence of two immobilized cations (ammonium in the side chain of lysine and guanidinium in the side chain of arginine) on the folding and assembly of melittin. Melittin adopts an α-helix structure and is driven by hydrophobic interactions to associate into a helical bundle. To test the influence of immobilized cations on the peptide structure, we designed the homozygous mutants exclusively containing ammonium (melittin-K) or guanidinium (melittin-R) and compared the differences of melittin-K vs. melittin-R in their folding, assembly, and molecular functions. The side chains of lysine and arginine differ in their influences on the folding and assembly of melittin. Specifically, the side chain of R increases the α-helical propensity of melittin relative to that of K, following an inverse Hofmeister series. In contrast, the side chain of K favors the assembly of melittin relative to the side chain of R in line with a direct Hofmeister series. The opposite regulatory effects of immobilized cations on the folding and assembly of melittin highlight the complexity of the noncovalent interactions that govern protein intermolecular architecture.

The Structural Understanding of Transthyretin Misfolding and the Inspired Drug Approaches for the Treatment of Heart Failure Associated With Transthyretin Amyloidosis.

Substantial controversies exist in the exploration of the molecular mechanism of heart failure (HF) and pose challenges to the diagnosis of HF and the discovery of specific drugs for the treatment. Recently, cardiac transthyretin (TTR) amyloidosis is becoming recognized as one of major causes of underdiagnosed HF. The investigation and modulation of TTR misfolding and amyloidal aggregation open up a new revenue to reveal the molecular mechanisms of HF and provide new possibilities for the treatment of HF. The aim of this review is to briefly introduce the recent advances in the study of TTR native and misfolding structures, discuss the correlation between the genotype and phenotype of cardiac TTR amyloidosis, and summarize the therapeutic applications of TTR structural stabilizers in the treatment of TTR amyloidosis-associated HF.

Plant-Derived Nanovesicles: A Novel Form of Nanomedicine.

The nanovesicles extracted from the plant and herbal decoctions are identified as a new class of nanomedicine. They are involved in interspecies chemical communication across the plant and animal kingdoms and display a therapeutic potential against a variety of diseases. Herein, we review the recent progress made in the medical applications of plant-derived nanovesicles in the aspects of anti-inflammation, anti-cancer, tissue regeneration, and modulating commensal microbiota. We further summarize the cellular and molecular mechanisms underlying the physiological functions of plant-derived nanovesicles. Overall, plant-derived nanovesicles provide an alternative to conventional synthetic drugs and present exciting opportunities for future research on disease therapy.