Machine Learning-Enabled Image Classification for Automated Electron Microscopy.

Traditionally, materials discovery has been driven more by evidence and intuition than by systematic design. However, the advent of "big data" and an exponential increase in computational power have reshaped the landscape. Today, we use simulations, artificial intelligence (AI), and machine learning (ML) to predict materials characteristics, which dramatically accelerates the discovery of novel materials. For instance, combinatorial megalibraries, where millions of distinct nanoparticles are created on a single chip, have spurred the need for automated characterization tools. This paper presents an ML model specifically developed to perform real-time binary classification of grayscale high-angle annular dark-field images of nanoparticles sourced from these megalibraries. Given the high costs associated with downstream processing errors, a primary requirement for our model was to minimize false positives while maintaining efficacy on unseen images. We elaborate on the computational challenges and our solutions, including managing memory constraints, optimizing training time, and utilizing Neural Architecture Search tools. The final model outperformed our expectations, achieving over 95% precision and a weighted F-score of more than 90% on our test data set. This paper discusses the development, challenges, and successful outcomes of this significant advancement in the application of AI and ML to materials discovery.

Simultaneously improving accuracy and computational cost under parametric constraints in materials property prediction tasks.

Modern data mining techniques using machine learning (ML) and deep learning (DL) algorithms have been shown to excel in the regression-based task of materials property prediction using various materials representations. In an attempt to improve the predictive performance of the deep neural network model, researchers have tried to add more layers as well as develop new architectural components to create sophisticated and deep neural network models that can aid in the training process and improve the predictive ability of the final model. However, usually, these modifications require a lot of computational resources, thereby further increasing the already large model training time, which is often not feasible, thereby limiting usage for most researchers. In this paper, we study and propose a deep neural network framework for regression-based problems comprising of fully connected layers that can work with any numerical vector-based materials representations as model input. We present a novel deep regression neural network, iBRNet, with branched skip connections and multiple schedulers, which can reduce the number of parameters used to construct the model, improve the accuracy, and decrease the training time of the predictive model. We perform the model training using composition-based numerical vectors representing the elemental fractions of the respective materials and compare their performance against other traditional ML and several known DL architectures. Using multiple datasets with varying data sizes for training and testing, We show that the proposed iBRNet models outperform the state-of-the-art ML and DL models for all data sizes. We also show that the branched structure and usage of multiple schedulers lead to fewer parameters and faster model training time with better convergence than other neural networks. Scientific contribution: The combination of multiple callback functions in deep neural networks minimizes training time and maximizes accuracy in a controlled computational environment with parametric constraints for the task of materials property prediction.

Monitored COVID-19 vaccine humoral response in immunocompromised solid organ transplant recipients.

The SARS-CoV-2 pandemic has resulted in rapid research and vaccine development to help curtail unchecked transmission. However, these studies cannot be applied as easily among every population, such as immunocompromised individuals. In this study, we observed the humoral response of 70 total heart and renal transplant patients to mRNA SARS-CoV-2 vaccinations to help further understand the effectiveness of vaccination in post-transplant patients following second or booster vaccinations. Antibodies were measured by bead technology to detect IgG, as well as IgG/IgM Rapid Cassette tests for confirmation. Immunocompromised patients had a noticeably lower humoral response than non-immunocompromised populations, with an even lower response among Black patients. Our findings also show for the first time various antibody responses to different motifs of the virus, with the lowest being against the S2 motif. A potential link between the duration of immunosuppression and vaccine response was also observed, where patients on immunosuppressants for longer had a stronger response to vaccination compared to recent transplant patients in our study. In addition, younger transplant recipients had a better humoral response to vaccination, and vaccine effectiveness was disproportionate between races. This finding reinforces the continuation of the guidelines for accelerated vaccination schedules for immunocompromised patients.

Insights into next generation sequencing guided antibody selection strategies.

Therapeutic antibody discovery often relies on in-vitro display methods to identify lead candidates. Assessing selected output diversity traditionally involves random colony picking and Sanger sequencing, which has limitations. Next-generation sequencing (NGS) offers a cost-effective solution with increased read depth, allowing a comprehensive understanding of diversity. Our study establishes NGS guidelines for antibody drug discovery, demonstrating its advantages in expanding the number of unique HCDR3 clusters, broadening the number of high affinity antibodies, expanding the total number of antibodies recognizing different epitopes, and improving lead prioritization. Surprisingly, our investigation into the correlation between NGS-derived frequencies of CDRs and affinity revealed a lack of association, although this limitation could be moderately mitigated by leveraging NGS clustering, enrichment and/or relative abundance across different regions to enhance lead prioritization. This study highlights NGS benefits, offering insights, recommendations, and the most effective approach to leverage NGS in therapeutic antibody discovery.

Improving deep learning model performance under parametric constraints for materials informatics applications.

Modern machine learning (ML) and deep learning (DL) techniques using high-dimensional data representations have helped accelerate the materials discovery process by efficiently detecting hidden patterns in existing datasets and linking input representations to output properties for a better understanding of the scientific phenomenon. While a deep neural network comprised of fully connected layers has been widely used for materials property prediction, simply creating a deeper model with a large number of layers often faces with vanishing gradient problem, causing a degradation in the performance, thereby limiting usage. In this paper, we study and propose architectural principles to address the question of improving the performance of model training and inference under fixed parametric constraints. Here, we present a general deep-learning framework based on branched residual learning (BRNet) with fully connected layers that can work with any numerical vector-based representation as input to build accurate models to predict materials properties. We perform model training for materials properties using numerical vectors representing different composition-based attributes of the respective materials and compare the performance of the proposed models against traditional ML and existing DL architectures. We find that the proposed models are significantly more accurate than the ML/DL models for all data sizes by using different composition-based attributes as input. Further, branched learning requires fewer parameters and results in faster model training due to better convergence during the training phase than existing neural networks, thereby efficiently building accurate models for predicting materials properties.

MPpredictor: An Artificial Intelligence-Driven Web Tool for Composition-Based Material Property Prediction.

The applications of artificial intelligence, machine learning, and deep learning techniques in the field of materials science are becoming increasingly common due to their promising abilities to extract and utilize data-driven information from available data and accelerate materials discovery and design for future applications. In an attempt to assist with this process, we deploy predictive models for multiple material properties, given the composition of the material. The deep learning models described here are built using a cross-property deep transfer learning technique, which leverages source models trained on large data sets to build target models on small data sets with different properties. We deploy these models in an online software tool that takes a number of material compositions as input, performs preprocessing to generate composition-based attributes for each material, and feeds them into the predictive models to obtain up to 41 different material property values. The material property predictor is available online at http://ai.eecs.northwestern.edu/MPpredictor.

Generative Adversarial Networks and Mixture Density Networks-Based Inverse Modeling for Microstructural Materials Design.

There are two broad modeling paradigms in scientific applications: forward and inverse. While forward modeling estimates the observations based on known causes, inverse modeling attempts to infer the causes given the observations. Inverse problems are usually more critical as well as difficult in scientific applications as they seek to explore the causes that cannot be directly observed. Inverse problems are used extensively in various scientific fields, such as geophysics, health care and materials science. Exploring the relationships from properties to microstructures is one of the inverse problems in material science. It is challenging to solve the microstructure discovery inverse problem, because it usually needs to learn a one-to-many nonlinear mapping. Given a target property, there are multiple different microstructures that exhibit the target property, and their discovery also requires significant computing time. Further, microstructure discovery becomes even more difficult because the dimension of properties (input) is much lower than that of microstructures (output). In this work, we propose a framework consisting of generative adversarial networks and mixture density networks for inverse modeling of structure-property linkages in materials, i.e., microstructure discovery for a given property. The results demonstrate that compared to baseline methods, the proposed framework can overcome the above-mentioned challenges and discover multiple promising solutions in an efficient manner.

Moving closer to experimental level materials property prediction using AI.

While experiments and DFT-computations have been the primary means for understanding the chemical and physical properties of crystalline materials, experiments are expensive and DFT-computations are time-consuming and have significant discrepancies against experiments. Currently, predictive modeling based on DFT-computations have provided a rapid screening method for materials candidates for further DFT-computations and experiments; however, such models inherit the large discrepancies from the DFT-based training data. Here, we demonstrate how AI can be leveraged together with DFT to compute materials properties more accurately than DFT itself by focusing on the critical materials science task of predicting "formation energy of a material given its structure and composition". On an experimental hold-out test set containing 137 entries, AI can predict formation energy from materials structure and composition with a mean absolute error (MAE) of 0.064 eV/atom; comparing this against DFT-computations, we find that AI can significantly outperform DFT computations for the same task (discrepancies of [Formula: see text] eV/atom) for the first time.

A pandemic-enabled comparison of discovery platforms demonstrates a naïve antibody library can match the best immune-sourced antibodies.

As a result of the SARS-CoV-2 pandemic numerous scientific groups have generated antibodies against a single target: the CoV-2 spike antigen. This has provided an unprecedented opportunity to compare the efficacy of different methods and the specificities and qualities of the antibodies generated by those methods. Generally, the most potent neutralizing antibodies have been generated from convalescent patients and immunized animals, with non-immune phage libraries usually yielding significantly less potent antibodies. Here, we show that it is possible to generate ultra-potent (IC50 < 2 ng/ml) human neutralizing antibodies directly from a unique semisynthetic naïve antibody library format with affinities, developability properties and neutralization activities comparable to the best from hyperimmune sources. This demonstrates that appropriately designed and constructed naïve antibody libraries can effectively compete with immunization to directly provide therapeutic antibodies against a viral pathogen, without the need for immune sources or downstream optimization.

Cross-property deep transfer learning framework for enhanced predictive analytics on small materials data.

Artificial intelligence (AI) and machine learning (ML) have been increasingly used in materials science to build predictive models and accelerate discovery. For selected properties, availability of large databases has also facilitated application of deep learning (DL) and transfer learning (TL). However, unavailability of large datasets for a majority of properties prohibits widespread application of DL/TL. We present a cross-property deep-transfer-learning framework that leverages models trained on large datasets to build models on small datasets of different properties. We test the proposed framework on 39 computational and two experimental datasets and find that the TL models with only elemental fractions as input outperform ML/DL models trained from scratch even when they are allowed to use physical attributes as input, for 27/39 (≈ 69%) computational and both the experimental datasets. We believe that the proposed framework can be widely useful to tackle the small data challenge in applying AI/ML in materials science.

Robust IgM responses following intravenous vaccination with Bacille Calmette-Guérin associate with prevention of Mycobacterium tuberculosis infection in macaques.

Development of an effective tuberculosis (TB) vaccine has suffered from an incomplete understanding of the correlates of protection against Mycobacterium tuberculosis (Mtb). Intravenous (i.v.) vaccination with Bacille Calmette-Guérin (BCG) provides nearly complete protection against TB in rhesus macaques, but the antibody response it elicits remains incompletely defined. Here we show that i.v. BCG drives superior antibody responses in the plasma and the lungs of rhesus macaques compared to traditional intradermal BCG administration. While i.v. BCG broadly expands antibody titers and functions, IgM titers in the plasma and lungs of immunized macaques are among the strongest markers of reduced bacterial burden. IgM was also enriched in macaques that received protective vaccination with an attenuated strain of Mtb. Finally, an Mtb-specific IgM monoclonal antibody reduced Mtb survival in vitro. Collectively, these data highlight the potential importance of IgM responses as a marker and mediator of protection against TB.

Highly versatile antibody binding assay for the detection of SARS-CoV-2 infection and vaccination.

Monitoring the burden and spread of infection with the new coronavirus SARS-CoV-2, whether within small communities or in large geographical settings, is of paramount importance for public health purposes. Serology, which detects the host antibody response to the infection, is the most appropriate tool for this task, since virus-derived markers are most reliably detected during the acute phase of infection. Here we show that our ELISA protocol, which is based on antibody binding to the Receptor Binding Domain (RBD) of the S1 subunit of the viral Spike protein expressed as a novel fusion protein, detects antibody responses to SARS-CoV-2 infection and vaccination. We also show that our ELISA is accurate and versatile. It compares favorably with commercial assays widely used in clinical practice to determine exposure to SARS-CoV-2. Moreover, our protocol accommodates use of various blood- and non-blood-derived biospecimens, such as breast milk, as well as dried blood obtained with microsampling cartridges that are appropriate for remote collection. As a result, our RBD-based ELISA protocols are well suited for seroepidemiology and other large-scale studies requiring parsimonious sample collection outside of healthcare settings.

CAR-NK Cells Effectively Target SARS-CoV-2-Spike-Expressing Cell Lines In Vitro.

Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) is highly contagious and presents a significant public health issue. Current therapies used to treat coronavirus disease 2019 (COVID-19) include monoclonal antibody cocktail, convalescent plasma, antivirals, immunomodulators, and anticoagulants. The vaccines from Pfizer and Moderna have recently been authorized for emergency use, which are invaluable for the prevention of SARS-CoV-2 infection. However, their long-term side effects are not yet documented, and populations with immunocompromised conditions (e.g., organ-transplantation and immunodeficient patients) may not be able to mount an effective immune response. In addition, there are concerns that wide-scale immunity to SARS-CoV-2 may introduce immune pressure that could select for escape mutants to the existing vaccines and monoclonal antibody therapies. Emerging evidence has shown that chimeric antigen receptor (CAR)- natural killer (NK) immunotherapy has potent antitumor response in hematologic cancers with minimal adverse effects in recent studies, however, the potentials of CAR-NK cells in treating COVID-19 has not yet been fully exploited. Here, we improve upon a novel approach for the generation of CAR-NK cells for targeting SARS-CoV-2 and its various mutants. CAR-NK cells were generated using the scFv domain of S309 (henceforward, S309-CAR-NK), a SARS-CoV and SARS-CoV-2 neutralizing antibody (NAbs) that targets the highly conserved region of SARS-CoV-2 spike (S) glycoprotein and is therefore more likely to recognize different variants of SARS-CoV-2 isolates. S309-CAR-NK cells can specifically bind to pseudotyped SARS-CoV-2 virus and its D614G, N501Y, and E484K mutants. Furthermore, S309-CAR-NK cells can specifically kill target cells expressing SARS-CoV-2 S protein in vitro and show superior killing activity and cytokine production, compared to that of the recently reported CR3022-CAR-NK cells. Thus, these results pave the way for generating 'off-the-shelf' S309-CAR-NK cells for treatment in high-risk individuals as well as provide an alternative strategy for patients unresponsive to current vaccines.

Phosphatidylserine-Targeting Monoclonal Antibodies Exhibit Distinct Biochemical and Cellular Effects on Anti-CD3/CD28-Stimulated T Cell IFN-γ and TNF-α Production.

Phosphatidylserine (PS)-targeting monoclonal Abs (mAbs) that directly target PS and target PS via ß2-gp1 (ß2GP1) have been in preclinical and clinical development for over 10 y for the treatment of infectious diseases and cancer. Although the intended targets of PS-binding mAbs have traditionally included pathogens as well as stressed tumor cells and its associated vasculature in oncology, the effects of PS-targeting mAbs on activated immune cells, notably T cells, which externalize PS upon Ag stimulation, is not well understood. Using human T cells from healthy donor PBMCs activated with an anti-CD3 + anti-CD28 Ab mixture (anti-CD3/CD28) as a model for TCR-mediated PS externalization and T cell stimulation, we investigated effects of two different PS-targeting mAbs, 11.31 and bavituximab (Bavi), on TCR activation and TCR-mediated cytokine production in an ex vivo paradigm. Although 11.31 and Bavi bind selectivity to anti-CD3/28 activated T cells in a PS-dependent manner, surprisingly, they display distinct functional activities in their effect on IFN-γ and TNF-ɑ production, whereby 11.31, but not Bavi, suppressed cytokine production. This inhibitory effect on anti-CD3/28 activated T cells was observed on both CD4+ and CD8+ cells and independently of monocytes, suggesting the effects of 11.31 were directly mediated by binding to externalized PS on activated T cells. Imaging showed 11.31 and Bavi bind at distinct focal depots on the cell membrane. Collectively, our findings indicate that PS-targeting mAb 11.31 suppresses cytokine production by anti-CD3/28 activated T cells.

Field evaluation of a prototype tuberculosis lipoarabinomannan lateral flow assay on HIV-positive and HIV-negative patients.

Detection of tuberculosis at the point-of-care (POC) is limited by the low sensitivity of current commercially available tests. We describe a diagnostic accuracy field evaluation of a prototype urine Tuberculosis Lipoarabinomannan Lateral Flow Assay (TB-LAM LFA) in both HIV-positive and HIV-negative patients using fresh samples with sensitivity and specificity as the measures of accuracy. This prototype combines a proprietary concentration system with a sensitive LFA. In a prospective study of 292 patients with suspected pulmonary tuberculosis in Uganda, the clinical sensitivity and specificity was compared against a microbiological reference standard including sputum Xpert MTB/RIF Ultra and solid and liquid culture. TB-LAM LFA had an overall sensitivity of 60% (95%CI 51-69%) and specificity of 80% (95%CI 73-85%). When comparing HIV-positive (N = 86) and HIV-negative (N = 206) patients, there was no significant difference in sensitivity (sensitivity difference 8%, 95%CI -11% to +24%, p = 0.4351) or specificity (specificity difference -9%, 95%CI -24% to +4%, p = 0.2051). Compared to the commercially available Alere Determine TB-LAM Ag test, the TB-LAM LFA prototype had improved sensitivity in both HIV-negative (difference 49%, 95%CI 37% to 59%, p<0.0001) and HIV-positive patients with CD4+ T-cell counts >200cells/µL (difference 59%, 95%CI 32% to 75%, p = 0.0009). This report is the first to show improved performance of a urine TB LAM test for HIV-negative patients in a high TB burden setting. We also offer potential assay refinement solutions that may further improve sensitivity and specificity.