Machine learning with deep neural networks has demonstrated high performance for high dimensionality prediction tasks across multiple domains with sufficient sample data. Cardiovascular disease is a pertinent public health issue that has the potential to be better understood and addressed via deep learning approaches. In this work, we study machine learning approaches toward diagnosing various forms of cardiovascular disease and predicting its biomechanical behavior across multiple scales.

We begin by training deep learning models for an initial classification objective in echocardiography, a ubiquitous imaging modality for cardiologists. For view classification, we are able to demonstrate physician-level performance. We then expand the work from a methods and clinical application perspective. We address the high cost of annotation in medical imaging by examining data-efficient supervised and semi-supervised algorithms. In addition, we expand our prediction tasks towards the ultimate goal of automated, accurate cardiovascular disease diagnosis by predicting left ventricular hypertrophy.

To understand the nature of cardiovascular disease and develop treatments, a close look at the underlying biomechanics is important. For atherosclerosis, a leading cause of morbidity and mortality, we bridge finite element methods and machine learning to predict arterial tissue stress. Likewise for cytoskeletal proteins, which are the structural building blocks of human biology and influence cardiovascular health, we develop graph neural network algorithms to predict force response and conformational dynamics in calponin homology domains.

Moreover, we hope to lay the groundwork to advance the intersection of machine learning, biomechanics, and cardiovascular disease.

Selective yet efficient transport between the cell nucleus and cytoplasm is critical to cellular function as the nuclear pore complex is a major point of regulation for gene expression, signal transduction, immune response, oncogenesis and viral propagation. The combined physical structure of the pore, biochemical interaction of transport factors with pore components and the presence of the cellular milieu create conditions under which selective and efficient nucleocytoplasmic transport can occur. In this dissertation, I explore the significance of structure and interaction on transport of globular proteins and polymeric mRNA cargo through the crowded pore. In the first part of this work, an agent based modeling software framework is developed and validated for accurately simulating discrete and stochastic reaction-diffusion systems. A simulation environment representing the structure of the nuclear pore complex along with rules for the dynamics of protein movement and interaction was created using in vivo and in vitro reported parameters. This setup was then used to perform in silico experiments on the role of pore-cargo affinity in optimizing transport efficiency. These experiments demonstrate the pore's sensitivity to cargo affinity in maintaining efficient transport and suggest that a higher affinity binding site at the side of the pore where transport is terminated increases efficiency by reducing futile shuttling of cargo complexes. In the final part of this dissertation, I extend my agent based modeling framework to look at aspects of mRNA export that have remained unaddressed in experimental works. Variations in the number and spacing of transport receptors bound to the mRNA are shown to play a critical role in transport efficiency. In these experiments, a single transport receptor at the 5' end appeared insufficient for facilitating export. Increasing transport receptor coverage along the length of the mRNA improved the chances of successful export. Additionally, it was observed that the presence of a transport receptor near either the 5' or 3' terminus is required for successful export as it likely promotes the emergence of a favorable threading conformation. Finally, it was observed that the use of a single fluorescent tag to track and report mRNA export time, as is standard in current experimental work, is likely to underestimate true transport times. These findings have implications in the design of targeted delivery and export of polymeric molecules into and out the nucleus.

Leukocyte-endothelial adhesion and firm adhesion are chief steps in leukocyte recruitment cascade and immune system response to inflammation. Irregularities in these processes have been related to several human diseases including arthritis, cancer, and atherosclerosis. Given the fundamental importance of integrin activation and Ca2+ signaling in adhesion assembly and strengthening, understanding the molecular mechanisms of kindlin-mediated integrin activation and STIM-mediated Orai gating have potentially important implications in designing new therapeutic interventions. This study employs molecular dynamics and bioinformatic techniques to unravel detailed molecular interactions in these two activation processes.

First, we analyzed the effect of kindlin cooperation with talin for integrin activation. Integrins are transmembrane proteins that mediate the signaling between the cytoplasm and extracellular matrix to promote cell adhesion. In order to initiate signaling, integrins must be activated from a low-affinity state to a high-affinity conformation for binding to extracellular molecules. Talin and kindlin are two cytoplasmic proteins that bind to integrin and modulate its affinity for extracellular ligands. Although the molecular details of talin-mediated integrin activation are known, the mechanism of kindlin involvement in this process remains elusive. Here, we carried out, for the first time, a comprehensive molecular dynamics study of integrin activation mediated through simultaneous interaction of talin and kindlin with its b subdomain. We showed that kindlin modifies the molecular mechanisms of inside-out activation by enhancing talin interaction with the membrane proximal region of integrin and decreasing the crossing angle between transmembrane helices of integrin, which eventually results in parallelization of integrin dimer.

Second, we investigated the molecular details of STIM-Orai interaction and its downstream consequences on the Orai channel gating. The immune response is triggered by a decline in Ca2+ concentration within the endoplasmic reticulum (ER) that is followed by the opening of the calcium release-activated channels (CRAC), thus increasing the intracellular Ca2+ concentration. Stromal interaction molecule (STIM) has been identified as a Ca+2 sensor in the ER activating the CRAC channel via physical interaction with Orai, the pore unit of the CRAC channel. In this study, we showed that the forces from STIM binding allosterically propagate through Orai helices, which lead to opening of the basic region of the pore. Moreover, we demonstrated important interactions that maintain the open structure of the Orai channel.

Overall, this study provides valuable insights into the molecular mechanisms that contribute to the activation of mechanosensitive proteins involved in leukocyte adhesion and firm adhesion.

The nuclear pore complexes (NPCs) are the sole channels known on the nuclear envelope in eukaryotes through which nucleocytoplasmic traffic is selectively and efficiently conducted. While the NPC has been the subject of extensive research for the past six decades and many of its structural, biochemical, and biophysical details are revealed, mechanistics of selective transport yet to be known. The main game-player in conducting the selective transport is a class of the NPC proteins rich in Phe-Gly (FG)-repeat domains that are localized mainly to the channel interior. FG-repeat domains are natively unfolded and thus belong to the family of intrinsically disordered proteins (IDPs). The conformational behavior of FG-repeats remains unknown because of lack of detailed information about their microdynamics, leaving plenty of room for speculation on how the selectivity barrier forms inside the NPC.

To tackle the microdynamics of FG-repeats, here I used polymer physics’ principles to develop a coarse-grained computational biophysical model, incorporating the full sequence of all amino-acids. The simulations were run under Brownian dynamics approach to generate long time-evolution of FG-repeat domains.

Under known physiological conditions and geometrical constraints, FG-repeats form a spatially nonuniform cohesive meshwork, percolating in radial and axial directions, with a dense hydrophobic zone in the middle and a low-density zone near the wall. The FG-meshwork is extremely dynamic, resembling a jerking plug with a fluctuating concentration in radial direction.

Being porous with the dominant pore sizes of 4 and 6 nm, this dynamic meshwork is permeable to the active cargos in a hydrophobic, and to a lesser extent, charge, stimuli-responsive manner, but strongly impermeable to inert cargos having the same size. An active cargo creates a big deformation inside the FG-meshwork, but because of rapid Brownian motions of the FG-repeats, it reconstructs itself in in a cargo size and shape dependent manner, suggesting the individual FG-repeats undergo reversible collapse. Significantly, the reconstruction process follows a saturating exponential pattern with rapid and slow phases. The characteristic time of reconstruction is a function of cargo size and shape, and is generally smaller for the elongated cargos compared to globular cargos having the same surface chemistry.

Next, I used computational microrheology via many-particle tracking without external probe to investigate the full mechanical spectrum of FG-repeats under different physical and geometrical conditions, including FG-repeats’ composition, FG-repeats’ length, geometrical confinement, shuttling cargo, and end-tethering. The results reveal that FG-repeats show a non-Newtonian behavior as manifested in their shear-thinning viscosity. The viscoelastic response of FG-repeats is strongly frequency-dependent, and is consistent with the function of the permeability barrier at different frequencies, or equivalently, at different timescales. At low frequencies, equivalent to timescale of nucleocytoplasmic transport, FG-repeats form a pseudo solid-like meshwork. At high frequencies, equivalent to the timescale of thermal diffusion of small molecules, FG-repeats behave like a predominantly viscous liquid.

The end-tethering is determined to be the most influential factor in shaping the mechanical spectrum of the FG-repeats. When the end-tethering is lifted and FG-repeats get free in the space, they invariably behave as a non-Newtonian viscous liquid over all frequencies. The other factors investigated are geometrical confinement, FG-repeats composition, i.e. hydrophobicity and charge content, length of FG-repeats, and shuttling cargo. Although these factors might shift the value of viscoelastic response in the frequency domain, they do not change the physics of the response, as end-tethering does. Comparing the viscoelastic response of FG-repeats with a general polymer melt show a remarkable consistency between FG-repeats and polymer melt over the range of frequencies accessible to the current model. Ultimately, the answer to the question of “do FG-repeats form gel?” is discussed in detail.

Bidirectional transport of molecules through the nuclear envelope (nucleocytoplasmic transport) is a vital, yet very complex process inside eukaryotic cells. This process is facilitated by thousands of nanoscale pores embedded in the nuclear envelope, named nuclear pore complexes (NPCs). The remarkable feature of NPC is that it facilitates transport in a fast, yet selective manner. Despite numerous studies, the underlying mechanism of NPC’s function is unknown. The intricate function of NPC cannot be fully captured via experimental techniques due to small scale and rapid movement of the FG Nucleoporins, which are intrinsically disordered proteins rich in phenylalanine-glycine repeats, responsible for facilitating the transport process.

In this dissertation, using a range of computational approaches, the role of FG Nups in the nuclear transport process is explored. FG Nups do not have a well-defined secondary structure and it is believed that their sequence composition and conformational ensemble are critical for their function in the NPC. In this dissertation, for the first time, specific sequence patterns in the charge distribution of FG Nups were identified that were not observed in other intrinsically disordered proteins. These patterns are extended sub-sequences that only contain positively charged residues, have low charge density, and are located toward the N-terminus of FG Nups. We named these evolutionarily conserved patterns like charge regions (LCRs).

Additionally, the role that LCRs play in the conformational ensemble and function of FG Nups was examined, using coarse-grained molecular dynamics simulations. Our simulations in multiple levels (single Nups, ring cross sections of NPC, and whole NPC) show that number of charged residues in the LCR impact the conformational ensemble of FG Nups and movement of the cargo complex. Our simulations also suggest that the number of charged residues in LCR can regulate the interaction of cargo complex with FG Nups. Since conformational ensemble of FG Nups and their interaction with cargo complex are two governing factors of the transport process, we suggest that LCRs, the unique and evolutionarily conserved features of FG Nups, are major regulators of the nucleocytoplasmic transport.

Integrins are Alpha-Beta transmembrane receptors that mediate cell-­matrix and cell-­cell adhesion, comprised of multi-­domain, massive ectodomains, single-­pass, transmembrane domains, and short, floppy cytoplasmic domains. They link the extracellular matrix or counter-­receptors on other cells with the contractile cytoskeleton, mediating the transduction of mechanochemical signals across the plasma membrane and playing critical roles in a host of cellular functions, such as migration, cell traction, motility, platelet aggregation, and leukocyte transmigration.

Functionally, integrins are switch-­like proteins that can take on at least three different functional states: inactive, active, and ligand-bound. Integrin function is dependent upon allosteric conformational changes in its structure. Integrins are by default in an inactive (low affinity) state and can be activated via interacting with cytoplasmic proteins (e.g. talin) and/or engaging with extracellular ligands (e.g. fibrinogen). Integrin activation triggered by a cytoplasmic signal is called inside-­out signaling, while outside-­in signaling is defined as ligand-­integrin binding followed by conformational changes in the transmembrane and cytoplasmic domains. Association of the integrin with the ligand induces quaternary changes in the integrin, leading to cell signaling and dynamic cell adhesion. However, atomistic details of these conformational changes as well as mechanisms of integrin clustering are not fully understood.

This study employs molecular dynamics techniques to provide detailed, mechanistic answers for a few key questions on integrin (Alpha)IIb(Beta)3 function, a ­platelet-specific integrin member that plays a critical role in thrombosis. It is highly debated whether integrin transmembrane domains form homo-­oligomers, leading to focal adhesion growth. This study suggests that homo-­ oligomerization of the Beta subunit potentially regulates integrin clustering, as opposed to the Alpha subunit, which appears to be a poor regulator for the clustering process. Two distinct hypotheses are proposed to explain the atomic mechanism of integrin activation and how conformational changes triggered by cytoplasmic/extracellular proteins are propagated across the integrin structure: The switch-blade and the deadbolt model. To reconcile these apparently-contradictory models for integrin activation, this work investigated the mechanism of integrin (Alpha)IIb(Beta)3 inside-­out activation triggered by interactions with the cytoplasmic protein talin, and its outside-­in activation as a result of exposure to the soluble RGD ligand. Finally, it was shown that the integrin Alpha subunit head domain regulates integrin-­ligand binding affinity indirectly via inducing conformational changes in a key metal ion binding site (named LIMBS) in the Beta subunit head domain. Hence, it was concluded that different ligand binding affinities of integrin (Alpha)IIb(Beta)3 and (Alpha)V(Beta)3 is attributed to the larger attraction between the (Alpha)V subunit head domain and the metal ion binding site LIMBS.