Machine Learning Pipeline Research Articles

Rapid Estimation of the Intermolecular Electronic Couplings and Charge-Carrier Mobilities of Crystalline Molecular Organic Semiconductors through a Machine Learning Pipeline.

Organic semiconductors (OSC) offer tremendous potential across a wide range of (opto)electronic applications. OSC development, however, is often limited by trial-and-error design, with computational modeling approaches deployed to evaluate and screen candidates through a suite of molecular and materials descriptors that generally require hours to days of computational time to accumulate. Such bottlenecks slow the pace and limit the exploration of the vast chemical space comprising OSC. When considering charge-carrier transport in OSC, a key parameter of interest is the intermolecular electronic coupling. Here, we introduce a machine learning (ML) model to predict intermolecular electronic couplings in organic crystalline materials from their three-dimensional (3D) molecular geometries. The ML predictions take only a few seconds of computing time compared to hours by density functional theory (DFT) methods. To demonstrate the utility of the ML predictions, we deploy the ML model in conjunction with mathematical formulations to rapidly screen the charge-carrier mobility anisotropy for more than 60,000 molecular crystal structures and compare the ML predictions to DFT benchmarks.

Comprehensive assessment of machine learning methods for diagnosing gastrointestinal diseases through whole metagenome sequencing data

ABSTRACT The gut microbiome, linked significantly to host diseases, offers potential for disease diagnosis through machine learning (ML) pipelines. These pipelines, crucial in modeling diseases using high-dimensional microbiome data, involve selecting profile modalities, data preprocessing techniques, and classification algorithms, each impacting the model accuracy and generalizability. Despite whole metagenome shotgun sequencing (WMS) gaining popularity for human gut microbiome profiling, a consensus on the optimal methods for ML pipelines in disease diagnosis using WMS data remains elusive. Addressing this gap, we comprehensively evaluated ML methods for diagnosing Crohn’s disease and colorectal cancer, using 2,553 fecal WMS samples from 21 case-control studies. Our study uncovered crucial insights: gut-specific, species-level taxonomic features proved to be the most effective for profiling; batch correction was not consistently beneficial for model performance; compositional data transformations markedly improved the models; and while nonlinear ensemble classification algorithms typically offered superior performance, linear models with proper regularization were found to be more effective for diseases that are linearly separable based on microbiome data. An optimal ML pipeline, integrating the most effective methods, was validated for generalizability using holdout data. This research offers practical guidelines for constructing reliable disease diagnostic ML models with fecal WMS data.

Multiparameter quantitative analyses of diagnostic cells in brain tissues from tuberous sclerosis complex.

The advent of high-dimensional imaging offers new opportunities to molecularly characterize diagnostic cells in disorders that have previously relied on histopathological definitions. One example case is found in tuberous sclerosis complex (TSC), a developmental disorder characterized by systemic growth of benign tumors. Within resected brain tissues from patients with TSC, detection of abnormally enlarged balloon cells (BCs) is pathognomonic for this disorder. Though BCs can be identified by an expert neuropathologist, little is known about the specificity and broad applicability of protein markers for these cells, complicating classification of proposed BCs identified in experimental models of this disorder. Here, we report the development of a customized machine learning pipeline (BAlloon IDENtifier; BAIDEN) that was trained to prospectively identify BCs in tissue sections using a histological stain compatible with high-dimensional cytometry. This approach was coupled to a custom 36-antibody panel and imaging mass cytometry (IMC) to explore the expression of multiple previously proposed BC marker proteins and develop a descriptor of BC features conserved across multiple tissue samples from patients with TSC. Here, we present a modular workflow encompassing BAIDEN, a custom antibody panel, a control sample microarray, and analysis pipelines-both open-source and in-house-and apply this workflow to understand the abundance, structure, and signaling activity of BCs as an example case of how high-dimensional imaging can be applied within human tissues.

Efficiently Mitigating the Impact of Data Drift on Machine Learning Pipelines

Despite the increasing success of Machine Learning (ML) techniques in real-world applications, their maintenance over time remains challenging. In particular, the prediction accuracy of deployed ML models can suffer due to significant changes between training and serving data over time, known as data drift. Traditional data drift solutions primarily focus on detecting drift, and then retraining the ML models, but do not discern whether the detected drift is harmful to model performance. In this paper, we observe that not all data drifts lead to degradation in prediction accuracy. We then introduce a novel approach for identifying portions of data distributions in serving data where drift can be potentially harmful to model performance, which we term Data Distributions with Low Accuracy (DDLA). Our approach, using decision trees, precisely pinpoints low-accuracy zones within ML models, especially Blackbox models. By focusing on these DDLAs, we effectively assess the impact of data drift on model performance and make informed decisions in the ML pipeline. In contrast to existing data drift techniques, we advocate for model retraining only in cases of harmful drifts that detrimentally affect model performance. Through extensive experimental evaluations on various datasets and models, our findings demonstrate that our approach significantly improves cost-efficiency over baselines, while achieving comparable accuracy.

Medical-informed machine learning: integrating prior knowledge into medical decision systems

BackgroundClinical medicine offers a promising arena for applying Machine Learning (ML) models. However, despite numerous studies employing ML in medical data analysis, only a fraction have impacted clinical care. This article underscores the importance of utilising ML in medical data analysis, recognising that ML alone may not adequately capture the full complexity of clinical data, thereby advocating for the integration of medical domain knowledge in ML.MethodsThe study conducts a comprehensive review of prior efforts in integrating medical knowledge into ML and maps these integration strategies onto the phases of the ML pipeline, encompassing data pre-processing, feature engineering, model training, and output evaluation. The study further explores the significance and impact of such integration through a case study on diabetes prediction. Here, clinical knowledge, encompassing rules, causal networks, intervals, and formulas, is integrated at each stage of the ML pipeline, resulting in a spectrum of integrated models.ResultsThe findings highlight the benefits of integration in terms of accuracy, interpretability, data efficiency, and adherence to clinical guidelines. In several cases, integrated models outperformed purely data-driven approaches, underscoring the potential for domain knowledge to enhance ML models through improved generalisation. In other cases, the integration was instrumental in enhancing model interpretability and ensuring conformity with established clinical guidelines. Notably, knowledge integration also proved effective in maintaining performance under limited data scenarios.ConclusionsBy illustrating various integration strategies through a clinical case study, this work provides guidance to inspire and facilitate future integration efforts. Furthermore, the study identifies the need to refine domain knowledge representation and fine-tune its contribution to the ML model as the two main challenges to integration and aims to stimulate further research in this direction.

PATH-05. MACHINE LEARNING-BASED CYTOLOGICAL ANALYSIS OF CEREBROSPINAL FLUID IN MEDULLOBLASTOMA

Abstract BACKGROUND Metastatic dissemination occurring via the cerebrospinal fluid contributes to the poor prognosis in many medulloblastoma patients. It has high prognostic relevance, and its accurate detection is critical for adequate therapy stratification. However, cytological examination is a difficult, time-consuming task and frequently unreliable. We, therefore, propose a machine-learning pipeline to solve this diagnostic challenge. METHODS A dataset of 303 digitized images of cerebrospinal fluid preparations from 81 medulloblastoma patients was assembled. Based on these data, over 50000 digitized objects, including more than 5000 patches with tumor cells, were classified by experts into 13 clinically relevant diagnostic categories and used for training. The classification was performed using convolutional neural networks, trained with nested-cross validation and different transfer learning strategies. RESULTS The classification into the 13 diagnostic categories was feasible, with an overall accuracy of 75% on the test set and a tumor-specific F1 score of 89%. Most misclassifications occurred between morphologically similar cell types, mainly lymphocytes and activated lymphocytes, monocytes and activated monocytes, as well as tumor cells and atypical cells with borderline cytomorphology. Grouping these cytologically and/or biologically closely related groups increased the overall accuracy to 91%. Pretraining on published cytological datasets of cerebrospinal fluid and bone marrow further increased classification accuracy. CONCLUSIONS Deep learning can reliably detect tumor cells in digitized cytological images from cerebrospinal fluid preparations in medulloblastoma. Implementing explanation methods, an easy-to-use user interface and validation on additional cohorts will further strengthen the robustness of the presented approach.

Prediction of Diabetes Mellitus using Artificial Intelligence Techniques

Diabetes Mellitus (DM) is a global health challenge, demanding proficient predictive models for early identification and intervention. This study adopts a comprehensive strategy for diabetes prediction with Machine learning algorithms, utilizing PIMA Indian diabetes dataset which encompasses clinical, demographic and lifestyle data. Employing techniques like Recursive Feature Elimination (RFE) and correlation analysis, the feature selection process identifies influential predictors, including glucose levels, Body Mass Index (BMI), Blood Pressure and diabetic history of family. A distinctive facet of this study involves integrating IBM Auto AI, automating the machine learning pipeline for tasks like feature engineering, hyperparameter tuning and model selection. Through comparative analysis, the research evaluates the efficiency and performance enhancements achieved through automation in contrast to manually-tailored models. Evaluation metrics encompass accuracy, precision, recall, and F1 score. Crossvalidation, particularly k-fold cross-validation, ensures model generalization to diverse subsets of the dataset. The research outcomes offer valuable insights into the optimal amalgamation of AI techniques for diabetes prediction, underscoring the significance of interpretability, performance, and automation in healthcare analytics. The proposed Methodology is evaluated with different classifiers with Auto AI and without Auto AI techniques. Using IBM Auto AI,Gradient boosting algorithm performed well with 84.4 % accuracy and Logistic Regression showed good accuracy of 84. 4% among conventional machine learning techniques without Auto AI using Pima Indian Diabetes Dataset.

Determining water and solute permeability of reverse osmosis membrane using a data-driven machine learning pipeline

Water (A) and solute permeability (B) are the key parameters characterizing the performance of reverse osmosis (RO) membranes. However, determining A and B requires multiple times of experiments that are time-consuming. Thus, developing a method to estimate A and B quickly has been encouraged. This study introduces a data-driven machine learning (ML) pipeline designed to predict A and B for RO membranes. The ML pipeline addresses all stages for estimating A and B, from data preprocessing to practical guidance on developed models. In particular, the membrane composition data, an overlooked essential component detailing A and B, was employed as features for the ML models. Sixteen ML models were tested, and categorical boost (CatBoost) and extremely randomized trees (ET) were selected as the best models for A and B, respectively. The selected ML models were retrained using the minimum yet effective features identified through feature importance analysis. Consequently, the retrained ML models exhibited high accuracy in predicting A and B (Radj,test2 = 0.925/MAEtest = 0.246 for A; Radj,test2 = 0.986/MAEtest = 0.069 for B), demonstrating the viability of a data-driven approach for parameter estimation. Moreover, this study highlights that RO membrane performance can be gauged effectively using membrane composition data and a limited set of operating conditions.

Visualization of incrementally learned projection trajectories for longitudinal data

Longitudinal studies that continuously generate data enable the capture of temporal variations in experimentally observed parameters, facilitating the interpretation of results in a time-aware manner. We propose IL-VIS (incrementally learned visualizer), a new machine learning pipeline that incrementally learns and visualizes a progression trajectory representing the longitudinal changes in longitudinal studies. At each sampling time point in an experiment, IL-VIS generates a snapshot of the longitudinal process on the data observed thus far, a new feature that is beyond the reach of classical static models. We first verify the utility and correctness of IL-VIS using simulated data, for which the true progression trajectories are known. We find that it accurately captures and visualizes the trends and (dis)similarities between high-dimensional progression trajectories. We then apply IL-VIS to longitudinal multi-electrode array data from brain cortical organoids when exposed to different levels of quinolinic acid, a metabolite contributing to many neuroinflammatory diseases including Alzheimer’s disease, and its blocking antibody. We uncover valuable insights into the organoids’ electrophysiological maturation and response patterns over time under these conditions.

Contrastive machine learning reveals in EEG resting-state network salient features specific to autism spectrum disorder

We explore the potential of the contrastive variational autoencoder to detect latent disorder-specific patterns in the network, analyzing functional brain networks in autistic individuals as the case. Autism spectrum disorder has long troubled medical practitioners, neurologists, and researchers. It is due to its extremely variable nature, both neurologically and behaviorally. Though machine learning has been in use to automate autism diagnosis, little has been done to delve into its intricacies. Here, we attempt to understand the neural mechanisms of autism spectrum disorder using contrastive variational autoencoder in conjunction with feature engineering. Our proposed methodology results in a physiologically interpretable classifier with a remarkable F1-score (up to 95%) and reveals a weak frontal lobe functional connectivity in the alpha band for children with autism spectrum disorder. Our study suggests an increased focus on efficient frontal lobe EEG sampling. Additionally, it highlights the importance of the proposed pipeline for understanding the underlying neural abnormalities in autism over the traditional machine learning pipeline. Thus, the obtained results have proven a contrastive variational autoencoder to be a promising approach for discovering latent patterns and features in complex networks.

TensorKrowch: Smooth integration of tensor networks in machine learning

Tensor networks are factorizations of high-dimensional tensors into networks of smaller tensors. They have applications in physics and mathematics, and recently have been proposed as promising machine learning architectures. To ease the integration of tensor networks in machine learning pipelines, we introduce TensorKrowch, an open source Python library built on top of PyTorch. Providing a user-friendly interface, TensorKrowch allows users to construct any tensor network, train it, and integrate it as a layer in more intricate deep learning models. In this paper, we describe the main functionality and basic usage of TensorKrowch, and provide technical details on its building blocks and the optimizations performed to achieve efficient operation.

Automated machine learning methodology for optimizing production processes in small and medium-sized enterprises

Machine learning can be effectively used to generate models capable of representing the dynamic of production processes of small and medium-sized enterprises. These models enable the estimation of key performance indicators, and are often used for optimizing production processes. However, in most industrial applications, modeling and optimization of production processes are currently carried out as separate tasks, manually in a very costly and inefficient way. Automated machine learning tools and frameworks facilitate the path for deriving models, reducing modeling time and cost. However, optimization by exploiting production models is still in infancy. This work presents a methodology for integrating a fully automated procedure that embraces automated machine learning pipelines and a multi-objective optimization algorithm for improving the production processes, with special focus on small and medium-sized enterprises. This procedure is supported on embedding the generated models as objective functions of a reference point based non-dominated sorting genetic algorithm, resulting in preference-based Pareto-optimal parametrizations of the corresponding production processes. The methodology was implemented and validated using data from a manufacturing production process of a small manufacturing enterprise, generating highly accurate machine learning-based models for the analyzed indicators. Additionally, by applying the optimization step of the proposed methodology it was possible to increase the productivity of the manufacturing process by 3.19 % and reduce its defect rate by 2.15 %, outperforming the results obtained with traditional trial and error method focused on productivity alone.

RADAR-Pipeline: Scalable Feature Generation for Mobile Health Data

Introduction & BackgroundRADAR-Pipeline is an open-source Python framework designed to simplify and enhance mobile health data analysis. It has been designed to efficiently read and process the large amount of data generated through the RADAR-Base platform. RADAR-base is a scalable, real-time streaming and analytics open-source platform to facilitate research access and customisation requirements. Studies using the Radar-base platform have collected fine-grained longitudinal data from wearables and phones. The data can potentially create multitudes of digital biomarkers, which can be used to inform us greatly about the disease condition. Due to the sheer size of the data, it can be difficult for researchers to read and process those data -- a common task is identifying useful features and common data processing/analysis steps previously used by the community. Up to now, these have been hand-crafted by individual data scientists, often lacking the capability to be easily reused by the community without author-specific knowledge. Furthermore, generating variables based on already established research on a larger scale can be challenging and could hinder replication. Hence, we have designed RADAR-Pipeline to help researchers overcome these challenges. It empowers them to create and share their data analysis and visualisation pipelines, fostering collaboration and knowledge sharing within the research community. Objectives & ApproachThe primary objective of RADAR-Pipeline is to offer researchers a user-friendly and powerful platform to develop and share their research. Researchers can build reusable analysis and visualisation pipelines to ensure consistent and reliable results. It simplifies big data analysis by leveraging Apache Spark to handle large and complex mobile health datasets efficiently. Researchers can also save time and effort by reusing and extending existing pipelines built by others. Finally, the RADAR-Pipeline promotes collaboration and recognition by allowing researchers to share their work through the RADAR-base Analytics Catalogue, making their pipelines citable and accessible to the wider research community. Whilst Radar-pipeline has been designed to read data from Radar-base, it can also be used to read data from any dataset which uses Hadoop Distributed File System (HDFS) file system namespace. Relevance to Digital FootprintsMobile health data is rich and valuable for understanding human behaviour and health. RADAR-Pipeline addresses the challenges associated with analysing large and complex mobile health datasets, enabling researchers to extract valuable insights that can be used to (1) Improve public health: By enabling efficient analysis of large-scale mobile health data, RADAR-Pipeline can contribute to research efforts aimed at improving population health outcomes and developing effective interventions; (2) Personalised healthcare: By facilitating the extraction of individual-level features from mobile health data, RADAR-Pipeline can seamlessly be integrated with Kafka data streams and machine learning pipelines to process the data in real-time, which can then be utilised to create more effective and targeted real-time interventions. (3) Promote reproducible research: The framework's emphasis on transparency and reproducibility in research aligns with the conference's focus on the responsible use of digital mobile health data. Conclusions & ImplicationsRADAR-Pipeline is a valuable tool for researchers, offering them the means to harness the potential of mobile health data. By adopting this framework, researchers can achieve efficient and scalable data analysis, thereby streamlining the extracting insights from digital footprints. This efficiency enables researchers to delve deeper into the data and uncover valuable patterns and trends. Furthermore, RADAR-Pipeline promotes collaboration and knowledge sharing within the research community. By providing a standardised framework for data analysis, RADAR-Pipeline facilitates collaboration among researchers, leading to the sharing of best practices and the dissemination of knowledge.

Online learning and continuous model upgrading with data streams through the Kafka-ML framework

A pipeline of constant data streams is being built by the Internet of Things (IoT) to monitor information about the physical environment. In parallel, Artificial Intelligence (AI) is constantly developing and enhancing industrial, economic, and academic endeavors as well as quality of life thanks to these IoT data. In streaming contexts, Kafka-ML is our open-source framework that enables the management of Machine Learning (ML) and AI pipelines over data streams. Accordingly, it simplifies the deployment of Deep Neural Networks (DNNs) in practical applications. Nonetheless, this framework did not support the possibility of carrying out an Online Learning (OL) process, which is needed when new data are continuously arriving, and the models need to adapt to them on the fly. In this work, we have extended our previous work, the Kafka-ML framework, to enhance the management of ML/AI pipelines with OL features to enable both ML/AI distributed and centralized models to learn indefinitely over time. These models are continuously upgraded thanks to a process where automatic and flexible inference is carried out when improvements in the model performance are achieved. This opens up a large number of new possibilities within different fields of application, development, and work under the premise of incremental learning with ML models such as Electrical Vehicles and Industry 5.0. We have validated these new features by adapting and deploying state-of-the-art DNN models in different online scenarios, for both single and distributed configurations. The results show the capability of Kafka-ML to execute effective online training processes for ML models, improving their performance over time as new data becomes available.

TYK2 as a novel therapeutic target in Alzheimer's Disease with TDP-43 inclusions.

Neuroinflammation is a pathological feature of many neurodegenerative diseases, including Alzheimer's disease (AD)1,2 and amyotrophic lateral sclerosis (ALS)3, raising the possibility of common therapeutic targets. We previously established that cytoplasmic double-stranded RNA (cdsRNA) is spatially coincident with cytoplasmic pTDP-43 inclusions in neurons of patients with C9ORF72-mediated ALS4. CdsRNA triggers a type-I interferon (IFN-I)-based innate immune response in human neural cells, resulting in their death4. Here, we report that cdsRNA is also spatially coincident with pTDP-43 cytoplasmic inclusions in brain cells of patients with AD pathology and that type-I interferon response genes are significantly upregulated in brain regions affected by AD. We updated our machine-learning pipeline DRIAD-SP (Drug Repurposing In Alzheimer's Disease with Systems Pharmacology) to incorporate cryptic exon (CE) detection as a proxy of pTDP-43 inclusions and demonstrated that the FDA-approved JAK inhibitors baricitinib and ruxolitinib that block interferon signaling show a protective signal only in cortical brain regions expressing multiple CEs. Furthermore, the JAK family member TYK2 was a top hit in a CRISPR screen of cdsRNA-mediated death in differentiated human neural cells. The selective TYK2 inhibitor deucravacitinib, an FDA-approved drug for psoriasis, rescued toxicity elicited by cdsRNA. Finally, we identified CCL2, CXCL10, and IL-6 as candidate predictive biomarkers for cdsRNA-related neurodegenerative diseases. Together, we find parallel neuroinflammatory mechanisms between TDP-43 associated-AD and ALS and nominate TYK2 as a possible disease-modifying target of these incurable neurodegenerative diseases.

Identification of Family-Specific Features in Cas9 and Cas12 Proteins: A Machine Learning Approach Using Complete Protein Feature Spectrum.

The recent development of CRISPR-Cas technology holds promise to correct gene-level defects for genetic diseases. The key element of the CRISPR-Cas system is the Cas protein, a nuclease that can edit the gene of interest assisted by guide RNA. However, these Cas proteins suffer from inherent limitations such as large size, low cleavage efficiency, and off-target effects, hindering their widespread application as a gene editing tool. Therefore, there is a need to identify novel Cas proteins with improved editing properties, for which it is necessary to understand the underlying features governing the Cas families. In this study, we aim to elucidate the unique protein features associated with Cas9 and Cas12 families and identify the features distinguishing each family from non-Cas proteins. Here, we built Random Forest (RF) binary classifiers to distinguish Cas12 and Cas9 proteins from non-Cas proteins, respectively, using the complete protein feature spectrum (13,494 features) encoding various physiochemical, topological, constitutional, and coevolutionary information on Cas proteins. Furthermore, we built multiclass RF classifiers differentiating Cas9, Cas12, and non-Cas proteins. All the models were evaluated rigorously on the test and independent data sets. The Cas12 and Cas9 binary models achieved a high overall accuracy of 92% and 95% on their respective independent data sets, while the multiclass classifier achieved an F1 score of close to 0.98. We observed that Quasi-Sequence-Order (QSO) descriptors like Schneider.lag and Composition descriptors like charge, volume, and polarizability are predominant in the Cas12 family. Conversely Amino Acid Composition descriptors, especially Tripeptide Composition (TPC), predominate the Cas9 family. Four of the top 10 descriptors identified in Cas9 classification are tripeptides PWN, PYY, HHA, and DHI, which are seen to be conserved across all Cas9 proteins and located within different catalytically important domains of the Streptococcus pyogenes Cas9 (SpCas9) structure. Among these, DHI and HHA are well-known to be involved in the DNA cleavage activity of the SpCas9 protein. Mutation studies have highlighted the significance of the PWN tripeptide in PAM recognition and DNA cleavage activity of SpCas9, while Y450 from the PYY tripeptide plays a crucial role in reducing off-target effects and improving the specificity in SpCas9. Leveraging our machine learning (ML) pipeline, we identified numerous Cas9 and Cas12 family-specific features. These features offer valuable insights for future experimental and computational studies aiming at designing Cas systems with enhanced gene-editing properties. These features suggest plausible structural modifications that can effectively guide the development of Cas proteins with improved editing capabilities.

A No-Code Automated Machine Learning Platform for the Energy Sector

The energy sector is increasingly leveraging machine learning (ML) to enhance efficiency, sustainability, and decision-making processes. However, the complexity of ML model development and deployment poses significant challenges, particularly in adapting to diverse data environments and operational requirements. This paper presents the development of a No-Code Automated ML (Auto-ML) platform specifically tailored for the energy industry. We introduce a framework that automates critical steps in the ML pipeline, including data preprocessing, feature engineering, model selection, and hyperparameter tuning, while integrating domain-specific knowledge to handle unique industry challenges such as fluctuating energy demands and regulatory compliance. The platform employs a modular architecture that allows for customization and scalability, facilitating its adaptation across various energy sub-sectors, including renewable energy, oil and gas, and power distribution. The results underline the platform's potential to democratize advanced analytical capabilities within the energy industry, making sophisticated data-driven insights accessible even to non-expert users. This paper contributes to the field by detailing the architectural considerations, implementation challenges, and practical impacts of Auto-ML in a sector critical to global infrastructure and economic stability.

Machine Learning with Confidential Computing: A Systematization of Knowledge

Privacy and security challenges in Machine Learning (ML) have become increasingly severe, along with ML’s pervasive development and the recent demonstration of large attack surfaces. As a mature system-oriented approach, Confidential Computing has been utilized in both academia and industry to mitigate privacy and security issues in various ML scenarios. In this paper, the conjunction between ML and Confidential Computing is investigated. We systematize the prior work on Confidential Computing-assisted ML techniques that provide i ) confidentiality guarantees and ii ) integrity assurances , and discuss their advanced features and drawbacks. Key challenges are further identified, and we provide dedicated analyses of the limitations in existing Trusted Execution Environment (TEE) systems for ML use cases. Finally, prospective works are discussed, including grounded privacy definitions for closed-loop protection, partitioned executions of efficient ML, dedicated TEE-assisted designs for ML, TEE-aware ML, and ML full pipeline guarantees. By providing these potential solutions in our systematization of knowledge, we aim to build the bridge to help achieve a much stronger TEE-enabled ML for privacy guarantees without introducing computation and system costs.

An interactive atlas of genomic, proteomic, and metabolomic biomarkers promotes the potential of proteins to predict complex diseases

Multiomics analyses have identified multiple potential biomarkers of the incidence and prevalence of complex diseases. However, it is not known which type of biomarker is optimal for clinical purposes. Here, we make a systematic comparison of 90 million genetic variants, 1453 proteins, and 325 metabolites from 500,000 individuals with complex diseases from the UK Biobank. A machine learning pipeline consisting of data cleaning, data imputation, feature selection, and model training using cross-validation and comparison of the results on holdout test sets showed that proteins were most predictive, followed by metabolites, and genetic variants. Only five proteins per disease resulted in median (min–max) areas under the receiver operating characteristic curves for incidence of 0.79 (0.65–0.86) and 0.84 (0.70–0.91) for prevalence. In summary, our work suggests the potential of predicting complex diseases based on a limited number of proteins. We provide an interactive atlas (macd.shinyapps.io/ShinyApp/) to find genomic, proteomic, or metabolomic biomarkers for different complex diseases.

Risk Factors for Perinatal Arterial Ischemic Stroke: A Machine Learning Approach.

Perinatal arterial ischemic stroke (PAIS) is a focal vascular brain injury presumed to occur between the fetal period and the first 28 days of life. It is the leading cause of hemiparetic cerebral palsy. Multiple maternal, intrapartum, delivery, and fetal factors have been associated with PAIS, but studies are limited by modest sample sizes and complex interactions between factors. Machine learning approaches use large and complex data sets to enable unbiased identification of clinical predictors but have not yet been applied to PAIS. We combined large PAIS data sets and used machine learning methods to identify clinical PAIS factors and compare this data-driven approach with previously described literature-driven clinical prediction models. Common data elements from 3 registries with patients with PAIS, the Alberta Perinatal Stroke Project, Canadian Cerebral Palsy Registry, International Pediatric Stroke Study, and a longitudinal cohort of healthy controls (Alberta Pregnancy Outcomes and Nutrition Study), were used to identify potential predictors of PAIS. Inclusion criteria were term birth and idiopathic PAIS (absence of primary causative medical condition). Data including maternal/pregnancy, intrapartum, and neonatal factors were collected between January 2003 and March 2020. Common data elements were entered into a validated random forest machine learning pipeline to identify the highest predictive features and develop a predictive model. Univariable analyses were completed post hoc to assess the relationship between each predictor and outcome. A machine learning model was developed using data from 2,571 neonates, including 527 cases (20%) and 2,044 controls (80%). With a mean of 21 features selected, the random forest machine learning approach predicted the outcome with approximately 86.5% balanced accuracy. Factors that were selected a priori through literature-driven variable selection that were also identified as most important by the machine learning model were maternal age, recreational substance exposure, tobacco exposure, intrapartum maternal fever, and low Apgar score at 5 minutes. Additional variables identified through machine learning included in utero alcohol exposure, infertility, miscarriage, primigravida, meconium, spontaneous vaginal delivery, neonatal head circumference, and 1-minute Apgar score. Overall, the machine learning model performed better (area under the curve [AUC] 0.93) than the literature-driven model (AUC 0.73). Machine learning may be an alternative, unbiased method to identify clinical predictors associated with PAIS. Identification of previously suggested and novel clinical factors requires cautious interpretation but supports the multifactorial nature of PAIS pathophysiology. Our results suggest that identification of neonates at risk of PAIS is possible.