Large Datasets Research Articles

9 Delving into molecular underpinnings of sarcomatoid/rhabdoid dedifferentioation in renal cancers using spatial profiling

Abstract Background Renal cell carcinomas (RCC) are characterized by their largely diverse clinical outcomes. Leveraging genomic diversity among individual tumors, our research during the past decade has concentrated on developing personalized medicine approaches for the most common and the most aggressive type of RCC, clear cell renal carcinoma (ccRCC). We generated the largest dataset of ccRCC including somatic genomic and clinical annotations for over 940 ccRCC patients, and developed a genomic classifier, based on mutational status of 12 RCC-relevant genes, which is able to stratify patients according to their risk of relapse after nephrectomy and/or death due to RCC. Furthermore, we noticed the presence of mesenchymal-like cellular phenotypes in tumors of high-risk patients, representsing a de-differentiation process, known as sarcomatoic or rhabdoid (S/R) de-differentation. Methods The current knowledge about molecular mechanisms that may drive S/R de-differentiation is primarily generated from molecular profiling of bulk tumors, collecting data from a mixture of cells that co-exist in the tumor milieu. Therefore, high-resolution studies that precisely define molecular characteristics of different cellular phenotypes associated with S/R features are missing. We have used spatial transcriptomic profiling to investigate molecular mechanisms that underline S/R de-differentationin RCC. While preserving tissue context, we applied spatial whole human transcriptome profiling to areas exhibiting S/R or clear cell phenotypes within the same tumor specimens to generate phenotype-specific transcriptome profiles. In addition, we applied whole-exome sequencing (WES) to independent areas that exhibit different differentiation phenotype withing same tumors. Results Pathway and network analysis of genesets with upregulation in each area have revealed meaningful differences in cellular pathways which are active in each phenotype. Specifically, we observed that dysregulation of extracellular matrix (ECM) is a hallmark of S/R areas within RCC tumors. Furthermore, WES of phenotypically-distinct areas within each tumor has shed new light son the genome evoltuion of S/R phenotypes. Conclusions S/R dedifferentiation in RCC tumors is characterized by specific genomic evolutionary patterns and substantial dysregulation of ECM components. DOD CDMRP Funding: yes

IDMap: Leveraging AI and Data Technologies for Early Cancer Detection

Cancer screening is vital in cutting mortality rates, and containing the impact of cancer in a worldwide basis. The current conventional detection techniques including imaging and biopsy though efficient are also characterized with drawbacks like; invasive, expensive, and inaccurate. This abstract will describe the new AI and data solution in the fight against early cancer detection, which presents a massive opportunity to improve accuracy, cut down the time that it takes to deliver a diagnosis, and bring quality health care to possibly millions of patients. ML and, in particular, DL are prospective in terms of decision making upon medical data including imaging, genomic sequences, and electronic health records to detect biomarkers of cancer in early stages. The statistics show that the AI-driven systems are capable to provide better diagnostic outcomes than conventional methods in some fields including mammography for breast cancer and CT for lung. Moreover, AI’s integration in genomic studies helps in determining Cancer related genes and biomarkers hence supporting precision medicine that adapts treatment to the specific genetic information of the patient. Apart from having outlets in AI, big data analytics, cloud computing, and IoT are equally important in early cancer detection as well. Big data analysis enables the analysis of large and complicated data sets with the aid of which one may identify inklings that may point towards possible early development of cancer. The use of cloud computing in health care mainly provides meaningful platforms for the storage and management of the large volumes of medical data in a way that allows improved efficiency and high levels of security. Wearable sensors collect data on different biomarkers throughout a patient’s body, and convey real-time information regarding whether the biomarkers’ levels are approaching cancerous state. Despite this great promise, there are various issues that have to be solved: data protection, privacy, and security, problems with the algorithms’ biases, and integration into practice. Ethical questions are generally important to tackle the uncertainty surrounding data and decision-making in clinical care using A I systems. The future trends in early cancer diagnostics will involve deeper integration of the approaches as AI and big data technology, which will enable more precise prevention and treatment. The applicability of this approach can also extend to the early identification of cancer, but also the prevention of its occurrence through proper intervention. In conclusion, conversing AI and data technologies will be useful for enhancing the efficiency of the early cancer diagnosis, and that is why the perspectives for the patients’ recovery and the further decrease in the mortality rates connected with cancer are rather promising. The approaches in this area remain informative developing technologies that are likely to be integrated into clinical work as the leading organizational models for the future of oncology and preventive health.

ScTab: Scaling cross-tissue single-cell annotation models

Identifying cellular identities is a key use case in single-cell transcriptomics. While machine learning has been leveraged to automate cell annotation predictions for some time, there has been little progress in scaling neural networks to large data sets and in constructing models that generalize well across diverse tissues. Here, we propose scTab, an automated cell type prediction model specific to tabular data, and train it using a novel data augmentation scheme across a large corpus of single-cell RNA-seq observations (22.2 million cells). In this context, we show that cross-tissue annotation requires nonlinear models and that the performance of scTab scales both in terms of training dataset size and model size. Additionally, we show that the proposed data augmentation schema improves model generalization. In summary, we introduce a de novo cell type prediction model for single-cell RNA-seq data that can be trained across a large-scale collection of curated datasets and demonstrate the benefits of using deep learning methods in this paradigm.

Assessing climate change vulnerability: A village level analysis of the Indian west coast

The Indian west coast is under constant threat from climate change-induced hazards. Various social, economic, and infrastructural disparities along the coast cause significant variations in climate vulnerability. Current literature assesses vulnerability either over (1) a large area with poor spatial resolution or (2) a local area with better spatial resolution. The former assessments provide more comprehensive and broad insights into large spatial trends of vulnerability, while the latter provide more accurate and specific inputs needed by the local governments for effective intervention. However, there is a lack of studies that assess vulnerability simultaneously at a high-resolution and over a large geographic area, due to inadequacies in existing methodologies and difficulty in data management and analysis. This is a key gap that we address in our paper. We assess climate vulnerability of the entire Indian west coast at the village level, and propose a novel machine-learning based methodology tailored for high-resolution assessment over large geographic areas. This helped us produce the first high-resolution (i.e. village-level) climate vulnerability map of the entire Indian west coast. We found that the state of Maharashtra has the highest number of vulnerable villages and the state of Kerala has the least number of vulnerable villages. We collate and utilize a large dataset of 112 indicators describing socioeconomic characteristics, infrastructure and availability of financial services, among other aspects, to obtain a comprehensive picture of vulnerability. We analyze geospatial trends and attribute high vulnerability to specific indicators, which will help in effective decision-making at the village level.

Taxi Travel Distance Clustering Method Based on Exponential Fitting and k-Means Using Data from the US and China

The taxi travel distance distribution can be used to forecast the origin and destination (OD) distribution of taxis and private cars. Most of the existing studies on taxi trip distributions have summarized a “low–high–low” trend and approached zero at both ends; however, they failed to explain the reason for this distance distribution. The key indicators and parameters identified by various researchers using big data for the same city and year typically differ, especially in terms of the mode and mean values of distance and time. This study uses New York yellow and green taxi data (a total of 417,018,811 data points) from 2017 to 2022, as well as data from China, to obtain a general law of the taxi travel distance distribution through an analysis of the relative distance and relative frequency. The travel mode was 0.54 times the relative distance, while the data tended towards zero at 2.0 times the relative distance. We verified the reliability of the research method based on reference and survey data. The results reveal the formation mechanism of the taxi travel distance distribution characteristics, which follow an exponential distribution. These laws can be used in the context of urban planning and transportation research. We propose a taxi form distance clustering method based on the k-means approach, chosen for its effectiveness on large datasets, interpretability, and alignment with our research objectives. This method provides visual results for the travel distance and accurate information for urban transportation planning and taxi services. The practical implications for policymakers, urban planners, and taxi services are discussed, demonstrating how the identified travel distance distribution laws can influence urban planning and taxi service optimization. Finally, the problems of data collection, cleaning, and processing are identified from the perspective of data statistics and analysis.

Women's health study (WENDY) - a protocol of a population-based study assessing gynecological and metabolic health in women in their mid-30s.

The Women's Health Study (WENDY) was conducted to improve insights into women's health and health burden. It provides a unique, comprehensive data source that can be broadly utilised to understand gynaecological symptoms, diseases, and their relation to metabolic and overall health more deeply in a population-based setting. The study was conducted in Finland from May 2020 to October 2022. It included 1,918 women (33-37 years old) who were born in Northern Finland between July 1985 and December 1987. Data collection comprised one three-to-four-hour study visit that included clinical measurements, biological samples, ultrasound examinations and an extensive questionnaire on gynaecological and reproductive history, physical and mental health, quality of life, lifestyles, current life situations, health awareness and opinions. The study also included a menstrual cycle follow-up and cognitive testing up to three months via a mobile application. Given that all participants' data can be linked to all Finnish national registers, and the NFBC participants' data can be linked to the birth cohort data set collected from gestational week 24 onwards, WENDY study forms one of the largest datasets worldwide to investigate gynecological and metabolic health burden in women.

A Parallel Approach to Enhance the Performance of Supervised Machine Learning Realized in a Multicore Environment

Machine learning models play a critical role in applications such as image recognition, natural language processing, and medical diagnosis, where accuracy and efficiency are paramount. As datasets grow in complexity, so too do the computational demands of classification techniques. Previous research has achieved high accuracy but required significant computational time. This paper proposes a parallel architecture for Ensemble Machine Learning Models, harnessing multicore CPUs to expedite performance. The primary objective is to enhance machine learning efficiency without compromising accuracy through parallel computing. This study focuses on benchmark ensemble models including Random Forest, XGBoost, ADABoost, and K Nearest Neighbors. These models are applied to tasks such as wine quality classification and fraud detection in credit card transactions. The results demonstrate that, compared to single-core processing, machine learning tasks run 1.7 times and 3.8 times faster for small and large datasets on quad-core CPUs, respectively.

How safe are wild-caught salmons exposed to various industrial chemicals? First ever in silico models for salmon toxicity data gaps filling

Salmons are crucial to ecosystems and economic activities like commercial fishing and aquaculture, while also serving as an important source of nutrients, underscoring their ecological significance and the need for sustainable management. To better understand the toxicity and biological interactions between the salmon and industrial chemicals in the aquatic environment, we utilized the ToxValDB database to develop first ever computational toxicity models for six salmon subspecies (covering Atlantic and Pacific salmon) across two genera, employing Quantitative Structure-Activity Relationship (QSAR) and quantitative Read-Across Structure-Activity Relationship (q-RASAR) methods. For three smaller datasets (Oncorhynchus nerka, Oncorhynchus keta, and Oncorhynchus gorbuscha), we created mathematical models using the entire datasets where QSAR models demonstrated superior statistical quality compared to q-RASAR. Conversely, the three larger datasets (Oncorhynchus kisutch, Oncorhynchus tshawytscha, and Salmon salar) were divided into training and test sets, the q-RASAR models yielded better results compared to QSAR models. Mechanistic interpretations of these models revealed that descriptors such as Burden eigenvalues (BCUT), autocorrelation of topological structure (ATSC), and molecular polarizability were significant predictors of toxicity. For instance, higher polarizability and certain topological features were associated with increased toxicity as per the developed models. Statistically superior models for each subspecies were used to predict the aquatic toxicity of 1085 untested organic chemicals for toxicity data gap filling and risk assessment considering the applicability domain (AD). These insights are pivotal for designing safer chemicals and emphasize the need for sustainable management of salmon populations.

Machine Learning Prediction of On/Off Target-driven Clinical Adverse Events.

Currently, 90% of clinical drug development fails, where 30% of these failures are due to clinical toxicity. The current extensive animal toxicity studies are not predictive of clinical adverse events (AEs) at clinical doses, while current computation models only consider very few factors with limited success in clinical toxicity prediction. We aimed to address these issues by developing a machine learning (ML) model to directly predict clinical AEs. Using a dataset with 759 FDA-approved drugs with known AEs, we first adapted the ConPLex ML model to predict IC values of these FDA-approved drugs against their on-target and off-target binding among 477 protein targets. Subsequently, we constructed a new ML model to predict clinical AEs using IC values of 759 drugs' primary on-target and off-target effects along with tissue-specific protein expression profiles. The adapted ConPLex model predicted drug-target interactions for both on- and off-target effects, as shown by co-localization of the 6 small molecule kinase inhibitors with their respective kinases. The coupled ML models demonstrated good predictive capability of clinical AEs, with accuracy over 75%. Our approach provides a new insight into the mechanistic understanding of in vivo drug toxicity in relationship with drug on-/off-target interactions. The coupled ML models, once validated with larger datasets, may offer advantages to directly predict clinical AEs using in vitro/ex vivo and preclinical data, which will help to reduce drug development failure due to clinical toxicity.

Large-scale statistical study of the dependence of retention index on heating rate in temperature-programmed gas chromatography

Retention indices are values that characterize the retention of a compound in gas chromatography. In practice, retention indices are often assumed to depend only on the structure of the molecule and the type of the stationary phase, but this approximation is incorrect. This study is devoted to studying the dependence of retention indices on the column heating rate in the linear temperature programming mode, using a large and diverse data set. In the NIST 20 database, most data records are recorded in this mode. For stationary phases based on poly(5%-diphenyl-95%-dimethyl)siloxane (5%-phenyl-PDMS), there is a high proportion of records with heating rates of 10-15 K/min. In practice, such a high heating rate is rarely used and the use of such data may cause errors. A search was made for groups of records that were taken from the same primary source, recorded for the same compound and the same stationary phase, but differing in a heating rate. For each of these groups, the value D, the angular coefficient (slope) of the dependence of the retention index on the heating rate, was calculated. This value can take both positive and negative values. The highest values and the greatest variation of D values are observed for polar stationary phases, but further consideration was performed for 5%-phenyl-PDMS due to its greater practical significance. For these stationary phases, the highest D values are observed for aromatic and polyaromatic molecules; oxygen-containing compounds, on the contrary, exhibit lower D values. Negative D values are observed for many trimethylsilyl derivatives. A data set of D values for 756 molecules was selected and published online. There is almost no correlation between D and the retention index, lipophilicity factor logP, and molecular weight. Significant correlations with the number of cycles, the number of rotatable bonds, and the number of aromatic atoms were observed. Linear equations quantitatively relating the molecular descriptors to the D value were constructed. A number of cycles and halogen atoms were shown to contribute positively to the D value, while a number of oxygen atoms and bonds subject to internal rotation contributed negatively. The strong influence of the values related to the conformational rigidity of molecules and the weak influence of polarity allow us to suppose that the entropic factor has a key influence on the D value. A simple empirical linear equation for estimating the value of D is derived and presented in this study. Several machine learning methods for predicting D are compared. The best results are shown by gradient boosting and a random forest. However, the random forest does not achieve high accuracy in predicting the retention indices themselves.

Lidar cloud and aerosol layer detection method based on point cloud filtering

A point cloud filtering method is presented for atmospheric layer detection from lidar data. The method involves rising edge event recognition based on a wavelet transform function. Density-based clustering was then utilized to separate the real boundary from the original noisy point clouds based on continuous distribution characteristics of cloud and aerosol layer. Tests were carried out to verify the performance of our algorithm with synthetic lidar signals with noise. The layer base detection error within ± 5 bins was achieved for signals with SNRs higher than 3. Even for SNRs higher than 1, high consistency was still observed between retrieved results with our method and a visual analysis. These results indicate that our algorithm is suitable for unsupervised detection with large time-series datasets, such as CALIOP.

The stroke outcome optimization project: Acute ischemic strokes from a comprehensive stroke center

Stroke is a leading cause of disability, and Magnetic Resonance Imaging (MRI) is routinely acquired for acute stroke management. Publicly sharing these datasets can aid in the development of machine learning algorithms, particularly for lesion identification, brain health quantification, and prognosis. These algorithms thrive on large amounts of information, but require diverse datasets to avoid overfitting to specific populations or acquisitions. While there are many large public MRI datasets, few of these include acute stroke. We describe clinical MRI using diffusion-weighted, fluid-attenuated and T1-weighted modalities for 1715 individuals admitted in the upstate of South Carolina, of whom 1461 have acute ischemic stroke. Demographic and impairment data are provided for 1106 of the stroke survivors from this cohort. Our validation demonstrates that machine learning can leverage the imaging data to predict stroke severity as measured by the NIH Stroke Scale/Score (NIHSS). We share not only the raw data, but also the scripts for replicating our findings. These tools can aid in education, and provide a benchmark for validating improved methods.

Pretraining of 3D image segmentation models for retinal OCT using denoising-based self-supervised learning

Deep learning algorithms have allowed the automation of segmentation for many biomarkers in retinal OCTs, enabling comprehensive clinical research and precise patient monitoring. These segmentation algorithms predominantly rely on supervised training and specialised segmentation networks, such as U-Nets. However, they require segmentation annotations, which are challenging to collect and require specialized expertise. In this paper, we explore leveraging 3D self-supervised learning based on image restoration techniques, that allow to pretrain 3D networks with the aim of improving segmentation performance. We test two methods, based on image restoration and denoising. After pretraining on a large 3D OCT dataset, we evaluate our weights by fine-tuning them on two challenging fluid segmentation datasets utilising different amount of training data. The chosen methods are easy to set up while providing large improvements for fluid segmentation, enabling the reduction of the amount of required annotation or an increase in the performance. Overall, the best results were obtained for denoising-based SSL methods, with higher results on both fluid segmentation datasets as well as faster pretraining durations.

Interactive molecular causal networks of hypertension using a fast machine learning algorithm MRdualPC

BackgroundUnderstanding the complex interactions between genes and their causal effects on diseases is crucial for developing targeted treatments and gaining insight into biological mechanisms. However, the analysis of molecular networks, especially in the context of high-dimensional data, presents significant challenges.MethodsThis study introduces MRdualPC, a computationally tractable algorithm based on the MRPC approach, to infer large-scale causal molecular networks. We apply MRdualPC to investigate the upstream causal transcriptomics influencing hypertension using a comprehensive dataset of kidney genome and transcriptome data.ResultsOur algorithm proves to be 100 times faster than MRPC on average in identifying transcriptomics drivers of hypertension. Through clustering, we identify 63 modules with causal driver genes, including 17 modules with extensive causal networks. Notably, we find that genes within one of the causal networks are associated with the electron transport chain and oxidative phosphorylation, previously linked to hypertension. Moreover, the identified causal ancestor genes show an over-representation of blood pressure-related genes.ConclusionsMRdualPC has the potential for broader applications beyond gene expression data, including multi-omics integration. While there are limitations, such as the need for clustering in large gene expression datasets, our study represents a significant advancement in building causal molecular networks, offering researchers a valuable tool for analyzing big data and investigating complex diseases.

BIDSAlign: a library for automatic merging and preprocessing of multiple EEG repositories.

This study aims to address the challenges associated with data-driven electroencephalographic (EEG) data analysis by introducing a standardised library called BIDSAlign. This library efficiently processes and merges heterogeneous EEG datasets from different sources into a common standard template. The goal of this work is to create an environment that allows to preprocess public datasets in order to provide data for the effective training of deep learning architectures.
Approach. The library can handle both BIDS (Brain Imaging Data Structure) and non-BIDS datasets, allowing the user to easily preprocess multiple public datasets. It unifies the EEG recordings acquired with different settings by defining a common pipeline and a specified channel template. An array of visualisation functions is provided inside the library, together with a user-friendly GUI to assist non-expert users throughout the workflow.
Main results. BIDSAlign enables the effective use of public EEG datasets, providing valuable medical insights, even for non-experts in the field. Results from applying the library to datasets from OpenNeuro demonstrate its ability to extract significant medical knowledge through an end-to-end workflow, facilitating group analysis, visual comparison and statistical testing.
Significance. BIDSAlign solves the lack of large EEG datasets by aligning multiple datasets to a standard template. This unlocks the potential of public EEG data for training deep learning models. It paves the way to promising contributions based on deep learning to clinical and non-clinical EEG research, offering insights that can inform neurological disease diagnosis and treatment strategies.

On the Theoretical Quantification of Radii of Atoms in Molecules.

Despite the fundamental importance of radii of atoms in molecules for numerous applications in physics and chemistry, comprehensive methods for their theoretical evaluation are still scarce. Here, we present quantum chemistry-based approaches for evaluation of radii of atoms in molecules and assess their robustness by studying the agreement of van der Waals and solvent-excluded surfaces constructed by them with reference molecular surfaces. By studying a large data set of 1235 molecules, it is shown that estimation of radii via effective and free atomic volumes can accurately take the dependence of atomic radii on the chemical environment into account.

Evaluation of an innovative, open-source and quantitative approach for the kinematic analysis of rock slopes based on UAV based Digital Outcrop Model: A case study from a railway tunnel portal (Finale Ligure, Italy)

This paper presents a novel, cost-effective method for a rapid and quantitative characterization of discontinuities in fractured rock cliffs using Unmanned Aerial Vehicle-derived Digital Outcrop Models (DOMs). The proposed workflow combines manual extraction of discontinuity data from the DOM with a kinematic analysis using an automated algorithm (ROKA), which upgrades and renews the traditional Markland's test. The case study, a designed railway tunnel portal in NW Italy (Finale Ligure) demonstrates the advantages and limitations of this approach compared to current methods. Manual mapping is compared with (semi-)automatic extraction of discontinuity data from the DOM and validated with field-based scan lines. The results show that in complex geological settings, manual mapping provides more statistically robust and geologically reliable measurements of discontinuity orientation, validated by the user's expertise. Traditional stereographic approaches in kinematic analysis are limited in characterizing spatially variable slopes and discontinuity networks, providing only generic estimations of unstable slope surfaces and discontinuity intersections. This limitation is overcome by new algorithms that incorporate 3D spatial relationships of both the discontinuity network and the rock slope, resulting in a more accurate and site-specific kinematic characterization of the rock cliff, emphasized by 3D visualizations of the critical planes and potential rock failure volumes. The new approach significantly impacts the time, effort, and cost of the engineering projects, allowing to quantitatively define the potential unstable areas with a fast analysis. The Authors stress the importance of integrating digital workflows with comprehensive field-based characterizations of the local litho-stratigraphic and structural setting to effectively utilize large datasets and partially automated procedures.

A high-density linkage map reveals broad- and fine-scale sex differences in recombination in the hihi (stitchbird; Notiomystis cincta).

Recombination, the process of DNA exchange between homologous chromosomes during meiosis, plays a major role in genomic diversity and evolutionary change. Variation in recombination rate is widespread despite recombination often being essential for progression of meiosis. One such variation is heterochiasmy, where recombination rates differ between sexes. Heterochiasmy has been observed across broad taxonomic groups, yet it remains an evolutionary enigma. We used Lep-MAP3, a pedigree-based software that is efficient in handling large datasets, to generate linkage maps for the hihi or stitchbird (Notiomystis cincta), utilising information from >36 K SNPs and 36 families. We constructed 29 linkage maps, including for the previously unscaffolded Z chromosome. The hihi is an endangered passerine endemic to Aotearoa New Zealand that is sexually dimorphic and exhibits high levels of sexual conflict, including sperm competition. Patterns in recombination in the hihi are consistent with those in other birds, including higher recombination rates in micro-chromosomes. Heterochiasmy in the hihi is male-biased, in line with predictions of the Haldane-Huxley rule, with the male linkage map being 15% longer. Micro-chromosomes exhibit heterochiasmy to a greater extent, contrary to that reported in other birds. At the intra-chromosomal level, heterochiasmy is higher nearer to chromosome ends and in gene-rich regions. Regions of extreme heterochiasmy are enriched for genes implicated in cell structure. This study adds an important contribution in assessing evolutionary theories of heterochiasmy and provides a framework for future studies investigating fine-scale heterochiasmy.

Embracing Large Natural Data: Enhancing Medical Image Analysis via Cross-Domain Fine-Tuning.

With the rapid advancements of Big Data and computer vision, many large-scale natural visual datasets are proposed, such as ImageNet-21K, LAION-400M, and LAION-2B. These large-scale datasets significantly improve the robustness and accuracy of models in the natural vision domain. However, the field of medical images continues to face limitations due to relatively small-scale datasets. In this article, we propose a novel method to enhance medical image analysis across domains by leveraging pre-trained models on large natural datasets. Specifically, a Cross-Domain Transfer Module (CDTM) is proposed to transfer natural vision domain features to the medical image domain, facilitating efficient fine-tuning of models pre-trained on large datasets. In addition, we design a Staged Fine-Tuning (SFT) strategy in conjunction with CDTM to further improve the model performance. Experimental results demonstrate that our method achieves state-of-the-art performance on multiple medical image datasets through efficient fine-tuning of models pre-trained on large natural datasets.

Machine Learning Allowed Interpreting Toxicity of a Fe-Doped CuO NM Library Large Data Set─An Environmental In Vivo Case Study.

The wide variation of nanomaterial (NM) characters (size, shape, and properties) and the related impacts on living organisms make it virtually impossible to assess their safety; the need for modeling has been urged for long. We here investigate the custom-designed 1-10% Fe-doped CuO NM library. Effects were assessed using the soil ecotoxicology model Enchytraeus crypticus (Oligochaeta) in the standard 21 days plus its extension (49 days). Results showed that 10%Fe-CuO was the most toxic (21 days reproduction EC50 = 650 mg NM/kg soil) and Fe3O4 NM was the least toxic (no effects up to 3200 mg NM/kg soil). All other NMs caused similar effects to E. crypticus (21 days reproduction EC50 ranging from 875 to 1923 mg NM/kg soil, with overlapping confidence intervals). Aiming to identify the key NM characteristics responsible for the toxicity, machine learning (ML) modeling was used to analyze the large data set [9 NMs, 68 descriptors, 6 concentrations, 2 exposure times (21 and 49 days), 2 endpoints (survival and reproduction)]. ML allowed us to separate experimental related parameters (e.g., zeta potential) from particle-specific descriptors (e.g., force vectors) for the best identification of important descriptors. We observed that concentration-dependent descriptors (environmental parameters, e.g., zeta potential) were the most important under standard test duration (21 day) but not for longer exposure (closer representation of real-world conditions). In the longer exposure (49 days), the particle-specific descriptors were more important than the concentration-dependent parameters. The longer-term exposure showed that the steepness of the concentration-response decreased with an increased Fe content in the NMs. Longer-term exposure should be a requirement in the hazard assessment of NMs in addition to the standard in OECD guidelines for chemicals. The progress toward ML analysis is desirable given its need for such large data sets and significant power to link NM descriptors to effects in animals. This is beyond the current univariate and concentration-response modeling analysis.