Applied Research

CRISPR technology has made a breakthrough by enabling precise and efficient gene editing capabilities. For any genomic region, a guide RNA of the complementary sequence can lead a Cas9 enzyme to cut at a specific location. But, the cutting efficiency varies between different guide RNAs, and higher efficiency is desired. In this proposal, we plan to take advantage of the most comprehensive dataset of guide RNA efficiencies covering 80,000 guide RNAs by three Cas9 variants. We plan to utilize the whole dataset, train a multi-variant model, and apply additional computational improvements, to better predict guide RNA efficiencies. The final outcome will enable more accurate prediction of guide RNA efficiencies in CRISPR-Cas9 experiments for improved design of guide RNAs.

Deep learning models lack robustness to imperceptible adversarial and to large corruption distortions in image, text, and tabular data. In preliminary work, we find gradient-free trained sign activation networks are more robust to adversarial attacks than state of the art methods. We propose to develop new gradient-free training algorithms for deep convolutional sign activation neural networks jointly in this research proposal.

Emerging global grand challenges include:  detecting and giving attribution to cyber-threats,  detecting and preventing disease in human populations; and revealing community structure in large social networks.  Unlike traditional applications in computational science and engineering,  solving these data science problems at scale often raises new challenges because of the sparsity and lack of locality in the data, the need for research on scalable algorithms, and the development of frameworks for solving these real-world problems on high-performance computers,  and for improved models that capture the noise and bias inherent in the torrential data streams.  This collaboration focuses on dynamic,  streaming graph algorithms that will improve real-world applications in cybersecurity,  health,  and social sciences. The PIs will design new algorithms and devise efficient and scalable implementations using graph abstractions.  Streaming data is represented by the addition and removal of edges from the graph.  This seed project will support a  new collaboration to further the theoretical basis on graph learning from these dynamic, streaming graphs and produce efficient practical implementations of the results.

Collective behavior is all around us, whether we are commuting to work, or observing nature, where birds, fish, ants, insects and other animals form groups that seem synchronized in motion. Understanding the dynamics and emerging behaviors of such large concentrations of entities is important in many domains, ranging from traffic, urban management, disease control, multi-agent robotics to biological systems. For example, accurate models of collective behavior allow organizations to produce better outcomes associated with crowds, from easing congestion and bottlenecks in cities, to implementing anti-contagion policies for slowing the growth of COVID-19.  This project’s goal is to design the next generation of collective behavior models. Currently, we are developing a novel data-driven framework to predict the dynamics of crowds, swarms, and bacteria. The theoretical basis for this framework is Koopman theory, which allows collective dynamics to be automatically inferred by neural networks. With such automatic inference, we can predict future collective behaviors. Such prediction capability is expected to increase our understanding of collective dynamics, with a positive impact on collective behavior modeling with applications in urban management, biology, and medicine.

This seed project will explore selected parameterized problems, and aim to design faster algorithms. We focus on problems whose objectives are centred at the detection of trees or dominating sets in a given graph, and where we believe that the design of novel algebraic methods may be the key.

Space weather is a term used to describe changing environmental conditions in the solar system caused by eruptions on the Sun's surface such as solar flares. Understanding and forecasting of solar eruptions is critically important for national security and for the economy since they are known to have adverse effects on critical technology infrastructure such as satellites and power distribution networks. Space weather analytics is an emerging interdisciplinary field, which aims to (i) understand the onset of solar eruptions and assess space weather effects on Earth through big solar and space data analysis, and (ii) perform near real-time long-term predictions of extreme space weather events including solar flares, coronal mass ejections and solar energetic particles as well as solar wind and geomagnetic storms by using advanced artificial intelligence (AI) techniques. This project aims to develop a suite of image processing, computer vision and deep learning AI models and tools for performing space weather analytics. These AI tools will be used to predict solar eruptions and space weather events as well as to identify, detect, trace, and track patterns such as magnetic flux, H-alpha fibrils, loops, jets, filaments, coronal holes, switchbacks, spikes, polarity inversion lines in solar and space data.

Drones (a.k.a. Unmanned Aerial Vehicle, UAV) are a promising option for fire surveillance and detection. They are already being used as remote sensors to detect and monitor fires, find casualties and victims, as well as to provide bird’s eye view situational information, in real-time, to firefighters. They present many advantages, such as being widely available and able to cover fairly large areas at a low-cost. Most research so far has focused on the technical feasibility of using drones in different contexts (e.g., navigation, network, communication) and on the information processing of swarm data (e.g., multiple drones providing different views of the scene), but not conceptualizing the drone as part of a human-machine collaborative team.

As drones become increasingly autonomous, it becomes critical to conceptualize these drones in the context of fire-fighting ecology, which includes humans, such as firefighters, victims, search and rescue teams, health workers, and also bystanders. This involves integrating the drone into a “smart” system that can rely on real-time sensors and human input as well as AI to support the decision-making process and facilitate effective communication. In this collaboration, we propose to focus on two aims. The first will map how drones are currently integrated in firefighting teams and model data processes to best integrate drones in this context. The second is to create different prototypes of drones as a novel communication system and test how civilians (e.g., victims of fire) would interact with drones in different situations. Since such systems do not currently integrate human-machine collaboration, this project will be the first step towards determining the feasibility of intelligent drones in AI-based firefighting systems and will ultimately lead to direct insights into development of drone-integrated firefighting systems.

Fundamental aspects of parameter estimation unidentifiability in the interplay of models and data: improving clinical trials through mathematical modeling and machine learning.

Beam induced radiation damage is a major limiting factor for structure determination by cryo-EM. Exposure to the electron beam leads to the dissociation of the molecular structures and beam-induced drift. Both effects result in a loss of resolution. To mitigate these effects, micrographs are collected as multi-frame radiation-dose fractionated movies. current image processing methods filter-out and average the effect of radiation damage across the entire field of view, rather than using it as source of information. As a result, the cryo-EM maps generated by traditional image processing methods are oblivious to the chemical properties of the sample. We propose a novel strategy to identify different moieties in the 3D electron density maps by using our recently developed adaptive morphological neural networks. We will achieve this goal by extracting the chemical information encoded in cryo-EM datasets as local differences in the response to radiation.

Computing shortest paths in networks is a fundamental network design problem that has attracted lots of attention by the research community for more than half a century. Not surprisingly, this problem has also a multitude of real-world applications. Some of the real-world applications require the computation of several shortest (or nearly shortest) paths with some degree of “independence” among the computed paths. This proposal studies a wide class of abstract problems that find partially independent shortest (or nearly shortest) paths, for various definitions of independence.

Conventional public shared-IP networks such as the Internet cannot meet big data transfer demands due to the sheer data volume and unpredictable network performance. High-performance Networks (HPNs) featuring high-speed links and bandwidth reservation presents a promising solution to this problem. However, due to the fast-changing dynamics and high complexity of big data transfer in HPNs, end-users have not seen the corresponding increase in transport performance, and these expensive network resources are still suffering from severe underutilization. This project brings together the expertise from both NJIT and BGU to develop an intelligent transport advising framework that employs data-driven and machine learning-based approaches to enable big data transfer with predictable optimal performance. This framework consists of two major technical components: i) stochastic approximation-guided transport profiling optimization, and ii) machine learning-assisted transport performance prediction. These components can be integrated into prevailing data transfer services for practical use.

The Basuray lab at NJIT, in collaboration with the Arnusch lab and Bashouti lab at BGU, is developing a novel Point-of-Care (POC) electrochemical-based mRNA sensor called ESSENCE that meets the ASSURED criteria for POC devices by WHO for the rapid detection of viruses. ESSENCE uses a Shear-Enhanced, flow-through non-planar 3D Nanoporous Electrode to overcome current electrochemical sensors' selectivity and sensitivity limitations for the rapid, sensitive, and selective detection of biological molecules DNA, liquid biopsy, and emerging contaminants like PFAS in different matrices. ESSENCE will be modified to pack tailored laser-induced graphene material from BGU to detect the target virus, Hepatitis C, and study matrix effects, obtain calibration curves, and determine the detection limit. ESSENCE will be extended to other viruses to develop a universal platform.

Water reclamation or wastewater reuse is critical to extend our limited water resources. However, the use of reclaimed wastewater effluents for agriculture is accompanied with concerns on the presence of emerging contaminants, like per- and polyfluoroalkyl substances (PFAS), and their incorporation in agricultural products. Further, the application of wastewater effluent and biosolid has been linked to an increase in the abundance of Antibiotic Resistance Genes (ARGs) in agricultural soils. However, questions remain unanswered on the role of plants and plant microbiome to control or dissipate ARG, in particular when exposed to emerging contaminants. Hence, this research aims to investigate the role of PFAS exposure on ARG expression by plant and plant microbiome using Lemna minor duckweed as a model aquatic plant. 

Recently, a large family of molecules called perfluoroalkyl substances (PFAS) have garnered attention as emerging contaminants of major concern as they can cause health problems in humans. However, because of their high stability and low concentration, traditional separation technology cannot efficiently remove them from aqueous environments; new materials and technologies are therefore urgently needed for remediation by capturing PFAS from the environment. In this project, a synergistic computational-experimental approach will be used to study PFAS removal from water using covalent organic framework (COF)-based membranes and identify promising COFs for PFAS remediation. High-throughput calculations and machine learning algorithms will be used to study PFAS interactions with COFs from a large database, while candidate COFs will simultaneously be synthesized and tested for PFAS removal.

In turbulent flows, particles of similar size tend to cluster together due to differences in inertia between the particles and the fluid they are suspended in. For drinking water treatment, the inertia difference is usually not sufficient to take advantage of this phenomenon for flocs. However, Drs. Katoshevski and Brenner have demonstrated that pulsed pressure waves can accomplish this same grouping phenomenon. Dr. Pennock will explore whether pulsed pressure waves can augment the performance of a pilot fluidized bed clarification system where turbulence is predicted to be too weak for the grouping phenomenon due to turbulence to be significant. Drs. Katoshevski and Brenner will simultaneously model the flow with computational fluid dynamics to compare with experimental results.

Textile-reinforced concrete (TRC) is a composite material that uses textile fabrics to improve the brittle tensile properties of concrete. Unfortunately, methods to improve the bond between textile fabrics and a concrete matrix are limited. This project will evaluate techniques to improve bond in TRC systems through a multi-scale approach using experimental, analytical, and numerical methods. These methods will generate understanding on the local behavior between a cementitious matrix and a textile fabric with the goal of creating more durable and resilient civil infrastructure. 

This project will develop and evaluate electrically assisted adsorption of a broad range of per- and polyfluoroalkyl substances (PFAS) using externally charged electrically conducting membranes (ECMs) made of selected carbonaceous nanomaterials (CNMs). The adsorption kinetics and capacity on selected conductive electrode material surfaces will be examined under variations of DC charges/currents. Meanwhile, the PFAS desorption kinetics on electrically conductive membranes under electrostatic control will be studied using environmentally relevant water chemistry. The overall goal is to assess the feasibility of concentrating PFAS via electrochemical membrane filtration and separate them for other chemical destruction disposal.