SolvedByAI

Queue

628 cases were harvested and are backed by a real paper, but they fail the test the index publishes on. Each one is listed here with the reason, so the exclusion can be checked rather than trusted.

All 628 assisted 358 not a result 183 unsupported 87
Paper Year Cited Why it is held
Deep learning-based framework for slide-based histopathological image analys... Scientific Reports 2022 44 assisted This is a methods paper proposing a new framework (HipoMap) and reporting its benchmark accuracy on existing, already-labeled tasks (TCGA lung cancer classification and survival analysis), so the AI's end-to-end predictions serve as validation of a technique rather than constituting the paper's contribution as a newly produced result. “HipoMap showed the Area Under the Curve (AUC) of 0.96±0.026 (5% improved) in the experiments for lung cancer classification”
Improved Weather Forecasting Using Neural Network Emulation for Radiation Pa... Journal of Advances in Modeling Earth Systems 2021 44 assisted The neural network replaces only the radiation parameterization, a single component, inside the human-built, human-run operational NWP pipeline at KMA, so the forecasts and their verification are products of a human workflow the AI merely accelerates rather than an end-to-end AI-produced result. “a neural network (NN) emulator for radiation parameterization was developed to use in an operational weather forecasting model in the Korea Meteorolog...”
Power System Transient Stability Assessment Based on Machine Learning Algori... Mathematics 2023 42 assisted The ML models are classification components developed and evaluated by humans within an emergency-control design workflow, with the abstract reporting benchmark accuracy on a test system (IEEE39) rather than the AI itself performing stability assessment or control end to end as the paper's solved result. “Algorithms were tested using the test power system IEEE39. In the case of the test sample, accuracy of instability classification for XGBoost was 91.5...”
Deep Learning-Based Attack Detection and Classification in Android Devices Electronics 2023 41 assisted The abstract reports a human-developed model's benchmark accuracy (precision, recall, efficiency) on existing malware datasets within a human-run pipeline of dataset creation, training, and evaluation, rather than an AI system itself producing the detection result end to end as the paper's contribution. “The effectiveness of our proposed model is validated using well-established datasets such as CICMalDroid2020, CICMalDroid2017, and CICAndMal2017.”
Bone mineral density response prediction following osteoporosis treatment us... Scientific Reports 2021 39 assisted The AI is explicitly a decision-support system that predicts BMD response to inform clinician treatment choices, with the 9.54% gain being a retrospective simulation rather than AI-driven treatment outcomes, so it aids rather than solves the personalized therapy task. “our novel approach using a machine learning-based decision support system is capable of predicting BMD response after osteoporosis treatment and perso...”
Diagnostic assessment of deep learning for melanocytic lesions using whole-s... Translational Oncology 2021 37 assisted The abstract explicitly positions the deep learning algorithm as a supplemental, pre-screening aid for pathologists rather than an agent that itself delivered the diagnostic outcome as the paper's contribution. “Deep learning algorithm might function as a supplemental tool to assist pathologist by automatically pre-screening and highlighting interest regions p...”
Intracranial Hemorrhage Detection in CT Scans using Deep Learning IEEE International Conference on Big Data Computing Service and Applications (BigDataService) 2020 36 assisted The model is explicitly framed as a tool to assist radiologists and accelerate their human-run diagnostic workflow, reporting benchmark accuracy on an existing RSNA dataset rather than the AI completing the clinical task itself. “To assist with this process, a deep learning model can be used to accelerate the time it takes to identify them.”
Classification of Dermoscopy Skin Lesion Color-Images Using Fractal-Deep Lea... Applied Sciences 2020 36 assisted The system is explicitly framed as a computer-aided diagnosis tool whose classification output supports dermatologists, and the paper reports benchmark accuracy (ISIC-2019) for a model operating as one component in a human clinical workflow rather than the AI itself producing a standalone solved result. “Computer-aided diagnosis is a helpful tool to help dermatologists in the detection of these kinds of illnesses.”
Genomic Prediction of Wheat Grain Yield Using Machine Learning Agriculture 2022 35 assisted This is a comparative evaluation of existing algorithms' prediction accuracy on an existing task (genomic prediction of wheat grain yield), where the models inform breeders rather than the paper's contribution being an AI-produced result. “we compared ML and deep learning (DL) algorithms with classical Bayesian approaches, across a range of different FS methods, for their performance in...”
Multi-Disease Detection in Retinal Imaging Based on Ensembling Heterogeneous... Studies in health technology and informatics 2021 35 assisted The paper presents a prediction pipeline whose output is explicitly framed as clinical decision support for human diagnosticians, with the contribution being benchmarked accuracy on existing retinal datasets rather than the AI itself performing diagnosis end-to-end. “Automated multi-disease detection models offer great potential to address this problem via clinical decision support in diagnosis.”
Prediction of evolutionary constraint by genomic annotations improves functi... Genome biology 2022 34 assisted The deep learning model predicts nucleotide conservation, but that prediction is one annotation component feeding a human-run pipeline of genomic prediction, experimental validation, and variant prioritization for breeding, rather than the AI itself producing the end result. “it improves genomic prediction for fitness-related traits such as grain yield, in elite maize panels, by stringent prioritization of fewer than 1% of...”
Antibiotic discovery with artificial intelligence for the treatment of Acine... mSystems 2024 33 assisted The AI (QSAR/virtual screening) only ranked candidate compounds, and the actual antibacterial result was produced and confirmed by human-run wet-lab assays (microdilution, time-kill, mutant studies), making AI one component in a human-completed discovery pipeline. “To validate these findings, one lead compound exhibiting promising binding stability as well as favorable pharmacokinetics properties, namely demethox...”
Detecting an axion-like particle with machine learning at the LHC Journal of High Energy Physics 2021 31 assisted The CNN is a tagging component inside a human-designed physics analysis whose contribution is projected sensitivity bounds on ALP couplings, not an AI-produced detection result. “With the CNN tagging algorithm, we demonstrate that our approach can extend current LHC sensitivity and probe the ALP mass range from 0.3 GeV to 5 GeV...”
MAGPIE: accurate pathogenic prediction for multiple variant types using mach... Genome Medicine 2024 31 assisted The abstract reports a machine-learning model's benchmark accuracy on an existing task (variant pathogenicity prediction against ClinVar-derived test and validation sets) and positions MAGPIE as a tool whose scores inform human variant interpretation, rather than an AI agent that end-to-end produced a solved primary result. “results underline the robustness of MAGPIE as a valuable tool for predicting pathogenicity in various types of human genome variations”
Deep learning-based prediction of in-hospital mortality for sepsis Scientific Reports 2024 30 assisted The DGFSD model is a clinical decision-support tool that reports superior accuracy on the existing task of sepsis mortality risk scoring, with its output explicitly framed as informing physicians who make the actual clinical decisions, so the AI assists rather than solves. “can help physicians in Intensive Care Unit make optimal clinical decisions”
Anti-Dengue: A Machine Learning-Assisted Prediction of Small Molecule Antivi... Viruses 2023 30 assisted The ML models only produced in-silico predictions of candidate inhibitors (unvalidated 'potential' repurposed drugs) within a human-built QSAR pipeline, rather than the AI itself solving the antiviral discovery problem end to end. “We identified goserelin, gonadorelin, and nafarelin as potential repurposed drugs with high pIC50 values.”
Machine Learning Approaches for Stroke Risk Prediction: Findings from the Su... Journal of Cardiovascular Development and Disease 2024 30 assisted Human researchers designed and ran this cohort analysis, employing standard ML models as analytical tools and reporting their performance metrics on the existing task of stroke risk prediction, with the model outputs (clusters, SHAP rankings) informing the authors' epidemiological conclusions rather than constituting an end-to-end AI-produced solution. “machine learning facilitated accurate stroke risk predictions and highlighted potential biomarkers, offering a data-driven framework for risk assessme...”
Enhancing PET Degrading Enzymes: A Combinatory Approach ChemBioChem 2024 29 assisted Machine learning is listed as one of several combined strategies (alongside rational design and ancestral sequence reconstruction) in a human-run enzyme engineering workflow that produced and validated the Combi-PETase variant, so the AI informed the design rather than producing the result itself. “we combine different strategies such as structure‐based rational design, ancestral sequence reconstruction and machine learning to engineer a more hig...”
Multi‐Agent‐Network‐Based Idea Generator for Zinc‐Ion Battery Electrolyte Di... Advanced Materials 2025 29 assisted The LLM network's role was limited to proposing/identifying candidate electrolyte compositions, while all actual discovery results (synthesis, conductivity, cycling performance, spectroscopic characterization) were produced by human-run wet-lab experiments, making this idea generation inside a human pipeline rather than the AI itself producing the result. “this work presents a Large Language Model (LLM)‐based multi‐agent network that proposes DEE compositions for zinc‐ion batteries”
Enhancing the Interpretability of Malaria and Typhoid Diagnosis with Explain... Tropical Medicine and Infectious Disease 2024 27 assisted The AI components (RF, LIME, GPT) produce predictions and explanations that inform healthcare workers who remain the diagnostic decision-makers, and the paper's contribution is an interpretability framework rather than an AI end-to-end solving of diagnosis. “This study employs explainable AI (XAI) models such as Local Interpretable Model-agnostic Explanations (LIME), and Large Language Models (LLMs) like G...”
Machine learning accelerated descriptor design for catalyst discovery in CO2... npj Computational Materials 2025 25 assisted The machine-learned force fields and unsupervised ML serve as a computational screening tool within a human-designed discovery framework, producing untested candidate predictions rather than a validated catalyst discovery. “we offer a powerful tool for catalyst discovery. We propose new promising candidates such as ZnRh and ZnPt3, which to our knowledge, have not yet been...”
Aboveground biomass estimation in a grassland ecosystem using Sentinel-2 sat... Environmental Monitoring and Assessment 2025 20 assisted This is a human-run remote-sensing study in which off-the-shelf RF and gradient boosting served as the regression component: the abstract reports model accuracy metrics on the existing task of AGB estimation rather than contributing an AI capability that performs the task end to end. “RF achieved the highest R2 value of 0.5755 Mg/ha, while gradient boosting achieved the highest R2 value of 0.7298 Mg/ha”
Deep Learning for Gravitational-Wave Data Analysis: A Resampling White-Box A... Sensors 2021 1 assisted The paper's contribution is a methodological resampling/white-box framework for understanding CNN uncertainties, with the CNN's detection performance reported as accuracy metrics on the existing GW detection task rather than the AI itself producing the scientific result. “we adopted a resampling white-box approach to advance towards a statistical understanding of uncertainties intrinsic to CNNs in GW data analysis”
Basset: learning the regulatory code of the accessible genome with deep conv... Genome Research 2016 1,047 assisted Basset is a predictive tool whose demonstrated contribution is accuracy on an existing prediction task and variant-annotation output that informs researchers' interpretation of the noncoding genome, rather than an AI-produced solution to a primary scientific problem. “Thus, Basset offers a powerful computational approach to annotate and interpret the noncoding genome.”
DeepCRISPR: optimized CRISPR guide RNA design by deep learning Genome biology 2018 634 assisted DeepCRISPR is a predictive design tool whose sgRNA efficacy/off-target predictions inform human-run CRISPR design workflows rather than the AI itself performing or producing the end biological result, and the abstract's claim is benchmark superiority on an existing in silico prediction task. “surpassing available state-of-the-art in silico tools”
CADD v1.7: using protein language models, regulatory CNNs and other nucleoti... Nucleic Acids Research 2023 440 assisted CADD is a variant-scoring tool whose predictions serve as one input to aid human clinical assessment and genetic studies, and the paper's contribution is improved benchmark performance of that predictive component rather than an AI end-to-end solving of a problem. “Machine Learning-based scoring and classification of genetic variants aids the assessment of clinical findings and is employed to prioritize variants...”
Brain age prediction using deep learning uncovers associated sequence varian... Nature Communications 2019 393 assisted The deep learning model produced brain age predictions that served as an intermediate phenotype for a human-run GWAS pipeline, meaning the AI's output was one component in a larger workflow rather than the end-to-end solution of the primary biological finding. “A genome-wide association study (GWAS) of PAD in the UK Biobank data... yielded two sequence variants”
Predicting protein–protein interactions through sequence-based deep learning Bioinformatics 2018 366 assisted The paper's contribution is a new prediction method whose primary result is benchmark accuracy on an existing task, with its predictions framed as computational tools to guide human discovery of PPIs rather than an end-to-end result produced autonomously by the AI. “Our experimental results show that DPPI outperforms the state-of-the-art methods on several benchmarks in terms of area under precision-recall curve (...”
Early detection of type 2 diabetes mellitus using machine learning-based pre... Scientific Reports 2020 366 assisted The paper is a comparative evaluation of ML models versus regression for T2DM screening within a human clinical workflow, not a demonstration that an AI system itself solved early diabetes detection; the models are prediction components whose outputs inform clinicians, and the contribution is the benchmark comparison concluding no clinically relevant improvement. “This study compares machine learning-based prediction models (i.e. Glmnet, RF, XGBoost, LightGBM) to commonly used regression models for prediction of...”
Optimized CRISPR guide RNA design for two high-fidelity Cas9 variants by dee... Nature Communications 2019 330 assisted The deep learning model is a gRNA activity predictor built from human-run genome-scale screens; it informs guide design rather than performing genome editing end-to-end, and the paper's contribution is a benchmarked prediction tool within a human workflow. “Our data reveals that a combination of RNN with important biological features outperforms other models for activity prediction. We further demonstrate...”
A unified connectomic target for deep brain stimulation in obsessive-compuls... Nature Communications 2020 316 assisted The connectivity-derived models served as an analytical component within a human-run workflow (surgery, cohort data collection, validation), informing researchers and surgeons rather than autonomously producing the clinical outcome, which came from human-performed DBS. “connectivity-derived models may inform clinical improvements across DBS targets, surgeons and centers”
Creating an automated trigger for sepsis clinical decision support at emerge... PLoS ONE 2017 308 assisted The ML model is a retrospective prediction component (a trigger) for sepsis clinical decision support, reporting AUC improvements to inform clinicians rather than performing the clinical task end to end. “utilizing free text drastically improves the discriminatory ability (increase in AUC from 0.67 to 0.86) of identifying infection”
The Rise of Data-Driven Weather Forecasting: A First Statistical Assessment... Bulletin of the American Meteorological Society 2024 211 assisted The paper's contribution is a human-run statistical verification of an existing ML model's (PanguWeather) forecast quality against an NWP system, an assessment of AI output rather than a result produced by the AI, with the AI serving only as the forecast-generating component in a human-completed evaluation. “we apply common forecast verification tools to assess to what extent a data-driven forecast produced with one of the recently developed ML models (Pan...”
Enhancing CRISPR-Cas9 gRNA efficiency prediction by data integration and dee... Nature Communications 2021 206 assisted CRISPRon is a prediction tool whose efficiency scores inform human gRNA design decisions (delivered as a design webserver), and the abstract reports benchmark accuracy of a model on an existing prediction task, making the AI a component of the human-run design workflow rather than the agent that produced the paper's primary result. “The design of CRISPR gRNAs requires accurate on-target efficiency predictions”
Deep learning for post-processing ensemble weather forecasts Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences 2021 196 assisted The deep neural network is a post-processing step layered on top of human-run numerical ensemble simulations, improving their output rather than producing the weather forecast end to end. “We propose a mixed model that uses only a subset of the original weather trajectories combined with a post-processing step using deep neural networks.”
Predicting gastric cancer outcome from resected lymph node histopathology im... Nature Communications 2021 186 assisted The deep-learning model performs tumor/lymph-node detection and computes the T/MLN ratio as a measurement component inside a human-designed study, with the prognostic-factor demonstration and cohort validation carried out by researchers, and the authors themselves frame the system as assisting pathologists and oncologists rather than producing the prognostic result end-to-end. “These findings indicate that deep-learning models could assist not only pathologists in detecting lymph nodes with metastases but also oncologists in...”
Machine learning in chemical reaction space Nature Communications 2020 184 assisted The ML provides learned reaction energies that inform human-driven extraction of reduced reaction networks and microkinetic analysis, within a workflow whose primary contribution is a newly established database and modeling framework rather than an AI-produced end result. “the learned reaction energies offer a non-empirical route to rationally extract reduced reaction networks for detailed microkinetic analyses”
A Deep Learning Approach Based on Explainable Artificial Intelligence for Sk... IEEE Access 2022 178 assisted The model is a decision-support tool whose benchmark performance (accuracy metrics on the existing ISIC 2019 task) is intended to inform dermatologists, who remain the agents completing the actual diagnosis. “This will help the dermatologists to make rational diagnosis in the early stages of skin cancer.”
Melanoma and Nevus Skin Lesion Classification Using Handcraft and Deep Learn... Entropy 2020 172 assisted The CNN serves only as a pre-trained feature extractor inside a human-designed CAD pipeline (preprocessing, handcrafted ABCD features, MI-based fusion, and separate classifiers), so the AI is one component of a workflow humans built and ran rather than the agent producing the classification result end to end. “deep learning features are extracted using a Convolutional Neural Network (CNN) architecture, which is pre-trained on Imagenet”
Interpretable Machine Learning for Early Prediction of Prognosis in Sepsis:... Infectious Diseases and Therapy 2022 159 assisted The paper reports a predictive model's accuracy (XGBoost, AUC 0.884) on an existing task (sepsis mortality prediction in MIMIC-IV) whose output is meant to inform clinicians rather than the AI itself producing the clinical result end-to-end. “the eXtreme Gradient Boosting (XGBoost) model achieved the best performance with an AUC of 0.884 and an accuracy of 89.5% in the validation cohort”