Our goal is to predictively engineer bio/nanomaterial hybrid systems with targeted functionality in a wide range of practical, technical, and medical applications. The open literature provides datasets of the functional properties of crystals, aqueous chemicals, and biological macromolecules, but the design of hybrid systems necessitates the modeling of all of these molecular species in a single common framework. Molecular graph convolutional networks and other deep learning methods are capable to train on datasets from multiple disciplines simultaneously, but in order to build these networks, a far-reaching data infrastructure is needed. We have created this infrastructure for three data sets: The Immune Epitope Database (IEDB) of MHC-I binding peptides, the Quantum-Machine.org QM9 dataset (QM9), and results extracted from the Materials Project. The IEDB provides binding affinities between biological macromolecules (peptide sequences in association with multiple MHC-I alleles); QM9 consists of 140,000 small organic molecules encoded as SMILES strings and 17 associated properties (including thermodynamic, energetic, geometric, and electronic information). The Materials Project dataset provides band gaps and formation energies for 70,000 crystal structures. We present a standardized train/test split and machine-learning-ready import interface for each of these datasets, as well as early results on co- and cross-training of deep neural networks across multiple datasets. The framework is expandable to new datasets and provides a strong foundation for ongoing efforts to build universal molecular encoding neural networks.

RNA interference (RNAi) therapy is a high-potential therapeutic solution that utilizes a short-interference RNA (siRNA) molecule to hinder the translation of a target gene. This is especially promising in the application of cancer therapy, as it can be used to suppress mutated oncogenic genes in tumor cells, presenting an alternative to conventional cancer therapy, such as chemotherapy and radiotherapy. However, delivering the siRNA successfully into tumor cells is challenging, as RNA are rapidly degraded in the blood serum. To address this, we developed an alternative siRNA delivery vector using chitosan-PEG coated iron-oxide nanoparticle (IONP:CP). The chemical flexibility of the polymer coating allows conjugation of various molecules, such as siRNA and targeting agents. Red fluorescent protein (RFP) was used as a reporter gene to demonstrate the gene knockdown mediated by the delivery of siRNA with IONP:CP. The siRNA was conjugated onto IONP:CP via disulfide bonds that can be cleaved in a reductive intracellular environment, allowing controlled release of the siRNA. Succinimidyl 3-(2-pyridyldithio) propionate (SPDP) assays and gel electrophoresis were performed to quantify successful siRNA conjugation onto the IONP:CP, while dynamic light scaterring (DLS) and zeta potentials were collected to characterize size and surface charge of the IONP. C6, a rat glioma cell line, transfected with the dsRed plasmid was utilized as our testing platform, and flow cytometry was used to measure fluorescence from the dsRed gene. A significantly low fluorescence levels in the cells would suggest successful siRNA silencing in the cell. We would also compare our data to other conventional methods of siRNA transfection, such as liposomal or polymer based transfection, to investigate the efficiency of our IONP vector.

The long term goal of this project is to study and mathematically characterize how the mammalian (mainly Homo sapiens) brain creates and maintains meaningful neuronal connections and organizes into connectomes and cortexes, and to compare what we learn to an existing (patented and under development at GEMSEC) brain-like computational neural network. A key point in understanding the formation, organization and long-term existence of new brain neural networks is the fundamental relationship between the geometric entropy of the physical network embedding, and information entropy of the network adjacency, connectivity. To our knowledge, there is currently no study in the literature that focuses on understanding biological neural networks through the entropy of their connections. Here we pose the question whether entropy related learning rules emerge from biological network connections or are the driving force for these connections. We also ask whether such learning rules can be imposed on artificial neural networks for enhanced functionality. To find the information entropy analogue of the geometric entropy term from a network point of view, we need to define the information entropy of the neuronal adjacency matrix. We define the information entropy of the neuronal adjacency, with random numbers between 0 and 1 symbolizing strength of the connections, i.e., synaptic plasticity, as the Shannon’s entropy, i.e., information entropy, of the spectral distribution of the neuronal adjacency matrix. To aid us in this study, two public neuronal datasets have been found: the Janelia FlyEM research group’s Hemibrain, and the NeuralEnsemble simulated spike train data. We test the entropy of the inferred neuronal connections from these datasets toward achieving our goal of the mechanism of formation of neural connections and connectomes. This project is supported by the UW Computational Neuroscience Center, and the DMREF Program of NSF through the MGI platform under DMR# 1629071, 1848911, and 1922020.

 Spontaneous self-organization of solid-binding peptides on single-layer atomic materials offers enormous potential in employing these systems for technological and medical applications from biosensors to logic devices. Molecular self-organization of peptides depends highly on their sequences, which affect their conformational behavior under aqueous conditions. Traditional ways of computationally studying the effect of mutations on the conformation states involves dimension reduction on cosine and sine transformed torsion angles, often represented as Ramachandran plots. Although these studies successfully cluster conformation states, they fail to intuitively characterize the effect of the point mutation(s) directly, necessitating further data analysis. Here, we apply Hilbert Space-Filling-Curve, HSFC, on the torsion angles and demonstrate intuitive visualization for the effect of point mutations on secondary structure dynamics along a reaction coordinate. We perform molecular dynamics simulations on graphene using the graphite-binding dodecapeptide, WT-GrBP5. The 12-AA long peptide was selected by directed evolution using M13 based phage display. The WT-GrBP5 is known to self-organize on graphene under low-neutral pH at room temperature. A rationally designed charge-neutral mutant, M9-GrBP5, assembles at a broader range of pHs widening the range of practical implementations of the peptide. The HSFC shows that the mutated amino acids in M9 do not correlate with the reaction coordinate of pH change, unlike that of WT. Understanding the effect of amino acid φ-ψ pairs that contribute to the changes in the peptide’s conformational space, with changing conditions, will help in analyzing effects of point mutations. The effect of the peptide’s conformational behavior on their self-organization propensities on surfaces would lead to the design of sequences that form soft bio/nano interfaces with controlled molecular interactions towards strategies for practical applications. The research was supported by the DMREF Program at National Science Foundation (NSF) through the MGI platform (Materials Genome Initiative) under grant numbers DMR# 1629071, 1848911, and 1922020.

The ease of data retrieval, analysis, and distribution can accelerate the pace of scientific research. A difference in philosophies and methodologies in the manner of data-collection and storage leads to a lack of semantic-consistency in naming-conventions across disciplines. Previous studies have opened this discussion within various research disciplines, however there has yet to be a study accomplished within a Convergence Science Lab such ours, Genetically Engineered Materials Science and Engineering Center (GEMSEC). Well-defined naming-conventions, clear-cut data standards, and set programmatic interfaces are required to provide improved data access to researchers and the public. This consistency is necessary to create a robust database with the ability to query large amounts of related information. The key point is that integrating design (of variables, research questions, etc.) with traditional, empirical research approaches in the natural sciences will result in more robust data and clearer analysis. Here we discuss the need for standardized protocol and metadata standards for data-collection and storage. We began by creating a data analysis and storage pipeline for two computationally involved experiments, Molecular Dynamics (MD) and Next Generation Sequencing (NGS), within GEMSEC. A database schema storing metadata associated with raw, cleaned, and analyzed files was created. We found homogenous collections of metadata with inconsistencies, especially in peptide naming conventions, between experiments. The target-database includes those from other labs some of which may store unconventional file types, use other naming schemes for key variables such as sequences, and have different standards for what metadata is stored and what is computationally retrieved at a later time. This schema is formed via interviewing various other researchers, determining similarities and differences within metadata, and creating a standard for all to use based on this collected information.

Engineered solid binding peptides can be used as molecular tools for a variety of applications, with our focus lying on bio/nano interfaces. Peptide-solid interactions involve surface phenomena such as binding, surface diffusion, and self-organization on atomically flat solids. Each of these phenomena requires the knowledge of peptide folding patterns, which are difficult to study both experimentally and computationally. Molecular dynamics, MD, has been used to computationally model peptide/solid interactions, but without information regarding the energy landscape of peptide conformations, the challenge of predictive design remains. While several methods exist for finding the energy landscapes for single peptide systems, currently no approach is capable of handling multi-peptide/surface systems. Here we use dynamic mode decomposition, DMD, to efficiently explore the energy landscapes of such systems aiming to find accurate linear approximations for predictive design of peptides at bio/nano interfaces. We make use of extended, multiresolution DMD to allow for analysis of various datasets with identical timesteps while minimizing impacts of statistical noise. It is anticipated that some descriptions of conformation will be better suited to describe peptide conformation energy landscapes than others; based on this premise, we examined various descriptions of peptide conformations through DMD. These descriptions include interatomic distances, adjacencies and peptide backbone torsion angles, alongside wavelet, Hilbert, Fourier, and Laplace transformations of select datasets. Through such analyses, it is becoming possible to describe how peptide conformations in multi-peptide/surface systems evolve through the energy landscape and settle into energy minima corresponding to stable conformations. These conformations can then be corroborated to peptide self-assembly on the surface using scanning probe microscopy techniques with sub-A resolutions for the development of predictive design platforms for future applications in biosensors, bioelectronics, and logic devices. 

Self-assembling peptides show great promise as an effective bottom-up technique that can be used to integrate biology and nanoelectronics. Such peptides can form a long-range ordered structure on the surface of Single Layer Atomic Materials (SLAM), such as graphene and transition metal dichalcogenides (e.g. molybdenum disulfide) through a chiral recognition mechanism. As graphene is highly sensitive to the adsorption of molecules, it is an ideal building block for biosensors. However, controlling the self-assembly of biomolecules to attain functional structures remains an engineering challenge. One fundamental question is how does the van der Waals (vdW) interaction between the graphene sheet and the adsorbing peptide, and the resulting strain field, affect the peptide binding footprint and the formation of an ordered phase. To investigate the phenomenon, I used Atomistic Finite Element Modeling (AFEM) to simulate the strain fields that arise during physical adsorption of a peptide to the surface of a graphene sheet. I then compared data from the AFEM simulation to experimental data collected using Scanning Tunneling Microscopy (STM) to determine whether the strain patterns correlate with the experimentally observed chiral assembly directions. Establishing a correlation between adsorption induced strain and chiral recognition would provide a more robust hypothesis concerning the optimal conditions for the generation of long-range ordered biomolecular structures on SLAMs; and, by extension, a step towards the realization of peptide-based biosensors.

Proteins are the workhorses in biology and their functions are affected by their amino acid domains in a modular fashion. When placed on solid surfaces, such as single layer graphene, certain short-sequence solid-binding peptides possess specific conformation propensities that aid in the self-organization into long range ordered peptide nanowires on atomically flat crystal lattices, such as single atomic layer solids. The self-assembly of nanowires on two-dimensional solids create electronic junctions through the biological doping of the underlying material and, e.g., by ordering the dipole moments. The utilization of these self-assembled nanowires with unique electrical capabilities may be a significant step towards the advancement of biologically compatible electronic devices such as biosensors. One of the key aspects of the design, namely the driving force for self-assembly of peptides on inorganic crystals, is currently unknown. To further understand the surface phenomena, our goal is to apply theoretical computational modeling to simulate the behavior of peptides and analyze how various conditions, such as temperature, pH, concentration and molecular conformations impact self-assembly. Peptides are computationally simple and only act based on local information such as the conformation of their immediate neighbors on a shared surface. Therefore, the self-assembly behavior can be simulated by fully distributed and asynchronous Markov chain algorithms, which have previously not been applied to peptide folding behavior interacting with a substrate. From these simulation algorithms, we determined the ideal physical conditions and the driving-forces for self-assembled peptide-based bioelectronic networks interfaced with single layer materials. Understanding how to better form these bioelectronic networks could constitute the critical foundation to advance bio-nanotechnology through the creation of biosensors or other biologically compatible electronic devices.

Back contact solar cells improve on the standard geometry by arranging both the positive and negative electrodes on the back of the device. This significantly improves efficiency by eliminating three critical challenges faced by typical solar cells: (1) shading of the active layer by top contacts, (2) the conflicting requirement for conductive yet transparent top contact materials, and (3) difficulty in printing metallic contacts onto the sensitive photoactive layer. Electrodeposition is a disruptive method that offers an alternative pathway for solution processing manufacturing, but is relatively unexplored as a method of fabricating perovskite back contact solar cells. This study focuses on achieving selective electrodeposition of high quality nanoscale electron and hole transport layers (tin and nickel oxides respectively) onto interdigitated silver collectors. In this investigation, a silver working electrode and a nickel or tin counter electrode against a silver/silver chloride reference electrode comprise a three-probe system in nickel nitrate and tin chloride electrolyte baths. The conductivity and carrier mobility of the electrodeposited transport layers are discussed as a function of salt and dopant molecule concentrations in the electrolyte bath. High carrier mobility layers are desirable for increased solar cell efficiency. The microstructure and thickness of the transport layer are discussed as a function of bath temperature, stirring speed, and the magnitude and duration of the supplied current density. The transport layers are characterized using optical profilometry and scanning electron microscopy. Diode devices are fabricated to characterize the electrical properties of the oxide layers. Based on these results, an optimal electrodeposition procedure is recommended for producing high quality nickel- and tin-based charge transport layers for application in back contact perovskite solar cells. Achieving this device geometry via electrodeposition enables the production of more efficient solar modules using high throughput manufacturing methods - bringing novel photovoltaic materials one step closer to widespread use.

 Loss of tooth mineral, demineralization, is the root cause of dental ailments - the most prevalent health problems affecting over 90 percent of Americans. These range from white spot lesions, the earliest sign of dental demineralization, to periodontal diseases, which can lead to more serious health issues. Current restorative treatments of tooth structure and function utilize synthetic materials, e.g. amalgam, glass ionomers, and particle reinforced resin composites that lead to deposited secondary precipitates. Although these common procedures are well-established and relatively effective, their durability is limited due to lack of structural and functional integration of deposited layer with the underlying tooth. GEMSEC labs have developed a proprietary technology dubbed “peptide-guided remineralization” which enables the formation of a new mineral with protein-derived peptides. Using this technology, the lab teams have successfully restored dental hard tissues via several case studies including enamel, cementum, dentin under in vitro and in vivo conditions. Translating this technology into a daily-use product, we developed a prototype, dental lozenges, designed to aid in enamel remineralization using a biomineralizing peptide, ADP5, derived from amelogenin, the key enamel protein. Herein we aim to refine the lozenge formulation through an iterative study for enhanced durable and whitening remineralized layer. Remineralization performance of different tablet formulations were tested in artificial saliva using extracted human enamel teeth. The samples were characterized using SEM showing that the current lozenge formulation creates a new mineral layer on enamel up to 2 µm in thickness. In summary, the new lozenge artificially regenerates lost enamel on the molecular level to treat tooth decay and erosion. Developed through a simple biomimetic methodology, this prototype lozenge could be mass fabricated for the consumer dental care market and expanded to include dental varnishes, gels, and pastes.

Photovoltaic devices are a crux of renewable energy generation. Organic photovoltaic devices build on this by being flexible and easily processable. This research project aims to produce photovoltaic wires and a general process for solution-based wire coating. A thin stainless-steel wire is coated with three layers: an electron transport layer, a photoactive polymer layer, and a hole transport layer. Then this wire is wrapped with a silver counter electrode. These wrapped wires will be coated in a UV curable polymer to protect the polymer coating from degradation. This final product, a solar power generating wire, will have its photoconversion efficiency tested. Currently, our project team is characterizing the initial coated wire using scanning electron microscopy and optical microscopy to determine the effectiveness of our coating method. To supplement this, we are researching how others have tackled characterizing thin coatings for objects with small surface areas. We are also working on designing processing improvements to the wire coating device both by investigating industry wire coating techniques. We hope that our designs are an example of a scalable solution processing method for organic photovoltaic wires. Our prototype device can be used for general wire coating applications and the current photovoltaic product is an example of a potential product. In the future, this device could be used to solution process more efficient organic photovoltaic wires by leveraging different polymers and polymerization techniques.

Some peptides are known to form stable secondary structures due to their occupation of lower energy states. These folded peptides theoretically have a greater information entropy upon folding, but this has not been experimentally proven. One such peptide, (LK)7, which reliably folds into an α-helix, is used as a case study here to prove that uncertainty in electron energy values increases upon formation of stable secondary structures. We use molecular dynamics, MD, simulation software from Schrodinger to create atom positional data trajectories over the evolution of (LK)7 from its extended to α-helical forms. Using Python and the SciPy ecosystem we create atom adjacency matrices of each frame of the trajectory and weight these matrices by the atoms’ respective counts of valence electrons. We then calculate and plot the information entropy and energy based on these valence electron adjacency matrices over the evolution of (LK)7. Moving forward we will also create trajectories using different data from the same MD simulation. One of these trajectories will involve weighting of atom adjacency matrices by electrons in orbitals not limited to the valence shell. Another will include the atom positional data of water molecules in the system. The last trajectory will use both modifications. Using these trajectories, we plan to experimentally prove that electron information entropy generally increases upon the folding of a peptide to a stable secondary structure. 

A graphene-binding dodecapeptide, WT-GrBP5, spontaneously self-organizes on single layer graphite, which leads to a change in the electronic properties of the single atomic layer solid substrate. The peptide-2D solid hybrid system has the potential for applications in bioelectronics and biosensors. Self-organization of peptides on substrate is highly dependent on the peptide’s sequence and its conformational behavior on surfaces. To understand the molecular footprint of the peptide on graphene, it is essential to know the functional domains of the peptide that contribute to its ability to self-assemble. Here, we use alanine scan on WT-GrBP5 to analyze the contribution each amino acid has on the overall conformational landscape of the peptide and its interactions with graphene. Alanine scanning is a technique in which amino acids are replaced with alanine, to determine each amino acid’s effect on the peptide’s dynamics and conformational stability. Alanine is primarily used due to its small size and tendency to follow conformational preferences of other amino acids in a given peptide’s sequence. We ran Metadynamics simulations of the peptide and its Alanine-mutants on graphene, in order to sample the energy landscapes of the peptides in the solution as well as on graphene. Understanding the effect of certain amino acids on the peptide’s ability to assemble is crucial for identifying the molecular footprint of the peptide on the surface and how this contributes to the new physics that develops at these hybrid bio/nano interfaces. Our overall goal is to develop a predictive design model for bio/nano-interfaces for medical and technological applications in the future.

Major Histocompatibility Complexes (MHC) are transmembrane proteins that utilize a selective binding domain to recognize peptide fragments in the cell environment and display these antigens on the cell surface. This selectivity of binding to different substrates is a feature that would be highly useful to mimic in the realm of genetically engineered peptide-solid surface binding, with broad implications applicable to engineered biomimetic systems. Our goal is to engineer selective binding biological molecules by mimicking the characteristics of the MHC-1 protein binding slot. The conventional approach to this problem applies directed evolution in a lab setting, selecting mutant MHCs with higher binding affinity against an antigen of interest. This method is resource and time intensive. Instead, we propose a machine learning approach to this problem via molecular graph convolutional neural networks (MGCNs) which are structured just like the connected atoms of input molecules. To explicitly model the MHC-peptide binding event as a graph, it is necessary to find computationally tenable representations of the MHC binding site. Prior attempts represented MHC binding alleles using only the critical contact residue positions of the MHC. This method omits the protein architecture, making it untenable as a graph encoding strategy. In this study, we evaluate alternate approaches to generate graph encodings of the two actively-binding alpha helices in the MHC-I complex and evaluate their performance on the task of predicting antigen binding affinity. We present an open Python toolset for generating graphs of MHC-I alpha helices and preliminary evaluation of their performance on a regression task on the Immune Epitope Database MHC-I dataset. The ability to generate de novo binding molecules for unique surfaces such as cancer surface proteins, viral spike proteins, metallic surfaces etc. has various use cases in: diagnostics, therapeutics, and engineered biomimetic systems.

We are working on designing, developing, and understanding of an interesting class of magnetic meta-materials comprising of nanomagnets. These nanomagnets when arranged on square tile lattice forms a two-dimensional artificial spin ice which are important for modeling pyrochlore spin ice systems. We are currently engaged developing recipes for preparing thermally active ASI system comprised of permalloy (NiFe alloys). This will allow the system to explore the magnetic phase space configuration more efficiently and to achieve the true ground state of these many body magnetic system compared to conventional field demagnetization. We fabricate permalloy nanomagnets arrays out of thin film permalloy films (of thicknesses 10- 20 nm) following nanolithography processes involving electron beam lithography, metal mask transfer and ion milling process. The devices are fabricated on specially chosen SiN substrates which acts as diffusion barrier while performing post thermal annealing process at high temperatures. The permalloy films are initially tested for different annealing temperatures to analyze for any changes in surface morphology and magnetic properties. From this, a suitable safe range of annealing temperatures are determined and the ASI arrays of different permalloy thickness are subjected to controlled annealing. Once an optimum annealing temperature is found for safe thermal annealing, we subjected devices with different permalloy thicknesses between 10-15 nm to thermal demagnetization. The magnetic configurations in devices subjected to thermal demagnetization are imaged using magnetic force microscopy to determine their equilibrium configuration attained during annealing. Was found that for this thickness range a perfect demagnetization of ASI arrays are obtained in a temperature range of 320-350°C.

 The process by which proteins and peptides render biological functions through molecular recognition and signal transduction. Solid-binding-peptides, SBP, utilize a similar process, e.g., in biomineralization or self-organization of solid surfaces, e.g., during interaction with single-layer materials, soft bio/nano interfaces. To de-novo design peptides that both bind and spontaneously self-assemble upon a 2D material, such as MoS2, we can adapt the Resonant Recognition Model (RRM) that assumes that the process of molecular recognition is a resonant interaction. The RRM is a process that takes residue-averaged potentials along a protein-sequence and uses Fourier analysis to transform them into constituent-frequencies that are associated with specific 3D structures of the active site of proteins. We adapt the procedure to short 12-AA long peptide. When multiple SBPs share a similar behavior, such as binding to MoS2, we can find the resonant-frequency that correlates with MoS2 binding functionality. From there, we predict new peptides that possess the resonant-frequency and test their predicted functionality for veracity. For the approach, we utilize an in-house developed dataset of several hundred thousand peptides (selected through next generation sequencing) that bind to MoS2 with varying strengths, so we can calculate their key resonant-frequencies in order to isolate which frequency is associated with the binding with MoS2. This information aids us in eliminating candidate resonant-frequencies from a well characterized peptide developed in our lab, that both binds and self-assembles on MoS2, M6-GrBP5. This allows us to narrow down which frequencies, and therefore which peptides are candidates for self-assembling on MoS2. The research is underway to verify these predictions towards developing a generalized model.