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
Design of target specific peptide inhibitors using generative deep learning... Nature Communications 2024 319 assisted The VAE generated candidate sequences, but the final inhibitors were produced through a human-run pipeline of Rosetta docking, MD filtering, and wet-lab experimental assays, making the AI one generative component rather than the agent that solved the design task. “Our method integrates a Gated Recurrent Unit-based Variational Autoencoder with Rosetta FlexPepDock for peptide sequence generation and binding affini...”
Deep neural network correlation learning mechanism for CT brain tumor detect... Neural Computing and Applications 2021 284 assisted The paper's contribution is a proposed learning mechanism (CLM) whose value is demonstrated through standard accuracy/precision/recall metrics on an existing CT brain tumor detection task, i.e., a model-performance report inside the broader human-run medical imaging workflow rather than the AI itself producing the solved result. “Results show that our CLM model is able to reach about 96% accuracy, and about 95% precision and recall.”
Combining mechanistic and machine learning models for predictive engineering... Nature Communications 2020 282 assisted The machine learning models provided design recommendations within a human-run metabolic engineering pipeline (genome-scale model target selection, library construction, biosensor screening), with humans building and validating the improved strains rather than the AI producing the result end to end. “the best machine learning-guided design recommendations improving tryptophan titer and productivity by up to 74 and 43%, respectively, compared to the...”
A machine learning Automated Recommendation Tool for synthetic biology Nature Communications 2020 277 assisted ART is a recommendation/prediction tool whose outputs guide human engineers who still design, build, and test the strains in wet-lab DBTL cycles, making the AI a component in a human-run pipeline rather than the agent that solved the bioengineering. “ART provides a set of recommended strains to be built in the next engineering cycle, alongside probabilistic predictions of their production levels.”
Microbiome meta-analysis and cross-disease comparison enabled by the SIAMCAT... Genome biology 2021 265 assisted The paper's primary contribution is a software toolbox, and the ML models within it served as components of a human-designed and human-run meta-analysis pipeline (data curation, augmentation strategy, biomarker interpretation) rather than autonomously producing the result. “we developed SIAMCAT, a versatile R toolbox for ML-based comparative metagenomics. We demonstrate its capabilities in a meta-analysis of fecal metagen...”
Air Quality Index and Air Pollutant Concentration Prediction Based on Machin... Applied Sciences 2019 226 assisted This is a human-run application/evaluation study in which researchers applied off-the-shelf algorithms (SVR, RFR) to two existing public datasets and reported accuracy metrics, so the AI served as a predictive component in a workflow the humans built, ran, and interpreted rather than being the agent that produced the paper's primary result. “We used support vector regression (SVR) and random forest regression (RFR) to build regression models for predicting the Air Quality Index (AQI) in Be...”
A deep-learning framework for multi-level peptide–protein interaction predic... Nature Communications 2021 222 assisted The abstract presents CAMP as a benchmark-evaluated prediction tool whose outputs are meant to inform human researchers conducting peptide drug discovery, rather than the AI itself producing a solved scientific result. “CAMP can serve as a useful tool in peptide-protein interaction prediction and identification of important binding residues in the peptides, which can...”
Accelerating Atomic Catalyst Discovery by Theoretical Calculations‐Machine L... Advanced Energy Materials 2020 165 assisted The primary results (HER screening, adsorption energetics, reaction pathways) come from human-run DFT calculations, with the ML model serving as a corroborating prediction component whose output 'indicates a similar result' and only informs guidelines for experimental design. “the machine learning (ML) technique bag‐tree approach is innovatively utilized based on the fuzzy model for data separation and converse prediction of...”
Accelerated enzyme engineering by machine-learning guided cell-free expressi... Nature Communications 2025 134 assisted The ML models are one component of a human-run platform (cell-free assembly, expression, functional assays) that predicts variants to guide enzyme engineering rather than producing the engineered enzymes end-to-end. “We developed a machine learning (ML)-guided platform that integrates cell-free DNA assembly, cell-free gene expression, and functional assays to rapid...”
A deep learning, image based approach for automated diagnosis for inflammato... Annals of Translational Medicine 2020 126 assisted The abstract reports a CNN classifier's accuracy on the existing human-performed task of diagnosing inflammatory skin diseases and explicitly frames the model as a 'diagnosis assistant' to help clinicians, so the AI informs clinical diagnosis rather than autonomously producing the solved result. “we developed an artificial intelligence dermatology diagnosis assistant (AIDDA)”
CXNet-m1: Anomaly Detection on Chest X-Rays With Image-Based Deep Learning IEEE Access 2018 122 assisted This is a methods paper reporting a new CNN architecture's benchmark metrics on the existing ChestX-ray14 task, positioning the model as a diagnostic aid to improve human work efficiency and diagnosis accuracy rather than autonomously solving the clinical problem. “The experimental results show that our light model CXNet-m1 with sin-loss function achieves better accuracy rate, recall rate, F1-score, and AUC value...”
Machine learning of material properties: Predictive and interpretable multil... Science Advances 2022 114 assisted The ML models are re-run on existing benchmark datasets to report accuracy comparable to prior kernel/neural approaches, with the actual scientific insights derived by humans interpreting model coefficients, so the AI is a component in a human-driven methodological demonstration rather than the producer of a new result. “We revisit material datasets used in several works and demonstrate that simple linear combinations of nonlinear basis functions can be created, which...”
An Internet of Things System for Underground Mine Air Quality Pollutant Pred... Sensors 2018 104 assisted The ANN in Azure ML is one component of a human-designed and human-operated IoT monitoring pipeline (sensors, PCA, base station), providing MEI predictions that inform mine safety rather than the AI itself being the paper's end-to-end solved task. “The results of PCA were fed into the ANN model in AML studio, which enabled the prediction of MEI.”
Machine learning-guided acyl-ACP reductase engineering for improved in vivo... Nature Communications 2021 96 assisted The ML system guided sequence design within an iterative human-run design-test-learn pipeline where researchers performed the wet-lab construction, screening, and characterization to ultimately produce the improved enzymes, so the AI was a component rather than the agent that produced the result. “Over the course of ten design-test-learn rounds, we engineer enzymes that produce over twofold more fatty alcohols than the starting natural sequences...”
Deep learning model-assisted detection of kidney stones on computed tomograp... International braz j urol 2022 94 assisted This is a retrospective diagnostic-accuracy study evaluating a deep learning model's performance metrics against radiologist readings on an existing clinical task, not a demonstration of the model autonomously solving the clinical problem end to end. “The accuracy rate, sensitivity, specificity, and positive and negative predictive values of the deep learning model were determined.”
Generalizable sgRNA design for improved CRISPR/Cas9 editing efficiency Bioinformatics 2020 89 assisted The ML algorithm is a prediction/design tool within the larger human-run CRISPR editing workflow; humans performed the wet-lab plasmid library experiments and the actual genome editing, with the AI only scoring sgRNA candidates. “Differential sequence and structural features among the most and least potent sgRNAs were then used to train a machine learning algorithm for assay de...”
Machine learning models for predicting post-cystectomy recurrence and surviv... PLoS ONE 2019 83 assisted The ML models report prediction accuracy (>70% sensitivity/specificity) as a benchmark within a human-driven clinical workflow, with outputs intended to inform clinicians' monitoring and treatment decisions rather than the AI autonomously performing the clinical task end to end. “Such predictions may inform patient monitoring schedules and post-cystectomy treatments.”
Re-Routing Drugs to Blood Brain Barrier: A Comprehensive Analysis of Machine... IEEE Access 2022 82 assisted The paper's contribution is a comparative performance analysis of existing ML models on existing BBB datasets, with the resulting permeability predictions only informing human drug-repurposing decisions rather than the AI accomplishing repurposing end to end. “This study comprehensively analyzes the performance of the well-known machine learning models on two different datasets to overcome dataset-related bi...”
Measuring the model risk-adjusted performance of machine learning algorithms... Financial Innovation 2022 77 assisted The paper's contribution is a human-designed risk-adjustment framework for evaluating ML models; NLP and the ML models are components used within that human-run workflow rather than an AI system producing the primary result itself. “we compute the weight of each category according to the intensity of their mentions, using natural language processing and a risk terminology based on...”
Tailoring industrial enzymes for thermostability and activity evolution by t... Nature Communications 2025 75 assisted The machine learning model serves as a predictive component within the human-driven iCASE enzyme-engineering strategy, informing variant design that humans then construct and validate experimentally across four enzyme types, rather than the AI itself producing the improved enzymes end-to-end. “this dynamic response predictive model using structure-based supervised machine learning is established to predict enzyme function and fitness”
Forecasting severe grape downy mildew attacks using machine learning PLoS ONE 2020 68 assisted The ML models produce risk forecasts as one component of a human-run epidemiological workflow whose actual goal, reducing fungicide treatments via decision rules, is a projected human management outcome informed by the AI's predictions, not a result the AI itself delivered. “Our results also show that the use of fungicide treatment decision rules that take into account local characteristics would reduce the number of treat...”
Diagnosing Heart Failure from Chest X-Ray Images Using Deep Learning International Heart Journal 2020 66 assisted The paper reports a model's classification accuracy (82%) on an existing human-labeled dataset and explicitly frames the AI as a decision-support tool for human diagnosticians rather than as the agent that solved the diagnostic problem. “Deep learning can thus help support the diagnosis of heart failure using chest X-ray images.”
A Low-Cost High-Performance Data Augmentation for Deep Learning-Based Skin L... BME Frontiers 2022 61 assisted The paper's contribution is a plug-and-play data augmentation technique validated through benchmark classification metrics on existing datasets, with the AI's outputs explicitly framed as informing clinicians in a screening workflow rather than the AI itself solving the clinical problem end to end. “which can be combined with any model through a plug-and-play mode”
Testing the reliability of an AI-based large language model to extract ecolo... npj Biodiversity 2024 52 assisted This is a human-designed evaluation benchmarking an LLM's extraction speed and accuracy against a human gold-standard reviewer, and the authors themselves conclude the AI's output still requires human quality assurance to complete the task, making the LLM a component in a human-run workflow rather than the agent that solved it. “additional quality assurance steps are required to ensure data integrity”
Skin Lesion Classification using Deep Learning Architectures International Conference on Communication System, Computing and IT Applications (CSCITA) 2020 49 assisted The deep learning models report classification accuracy on an existing task and are explicitly framed as a pre-diagnosis aid whose output informs doctors and patients, who make the actual medical decision. “This can give the patients as well as doctors a good idea of whether or not there is a need for medical attention and can avoid unnecessary panic/fals...”
War and Deforestation: Using Remote Sensing and Machine Learning to Identify... Land 2023 49 assisted The LandTrendr algorithm and Tasseled Cap transformation were processing steps inside a human-designed remote sensing study; the researchers ran the workflow, interpreted the drivers of deforestation, and drew the policy conclusions, so AI/ML was a component rather than the agent that produced the result. “Spectral information was synthetized using the Tasseled Cap transformation and the time series data was simplified and despiked using the LandTrendr a...”
Machine learning models for prediction of adverse events after percutaneous... Scientific Reports 2022 48 assisted The paper reports comparative accuracy metrics of ML prediction models against existing risk scores to inform clinical decision-making, rather than an AI system performing the task end-to-end as the contribution. “To determine whether machine learning (ML) techniques predict peri-PCI adverse events ... with better discrimination or calibration than the National...”
A novel strategy for classifying the output from an in silicovaccine discove... BMC Bioinformatics 2013 46 assisted The ML system is a filtering/classification step within a human-run vaccine discovery pipeline, and the abstract explicitly states that true validation of candidates can only happen in the laboratory, so the AI informs rather than solves the problem. “Vaccine candidates from an in silico approach can only be truly validated in a laboratory.”
Effectiveness of Transfer Learning for Deep Learning-Based Electrocardiogram... Healthcare Informatics Research 2021 43 assisted The paper's contribution is a human-designed methodological evaluation showing that transfer learning improves ECG classifier F1-scores under data scarcity, with the deep learning models serving as instruments in a comparison experiment rather than being the agent whose output constitutes the result. “Transfer learning effectively overcomes the data shortages that can compromise ECG domain analysis by deep learning.”
Motion2Vec: Semi-Supervised Representation Learning from Surgical Videos Proceedings - IEEE International Conference on Robotics and Automation 2020 43 assisted The paper's contribution is a human-designed representation-learning algorithm (Motion2Vec) whose merit is reported as benchmark accuracy on the existing JIGSAWS task against state-of-the-art baselines, rather than an AI-produced result that itself constitutes the contribution. “Results give 85.5% segmentation accuracy on average suggesting performance improvement over several state-of-the-art baselines, while kinematic pose i...”
Protein–Protein Interactions Prediction via Multimodal Deep Polynomial Netwo... IEEE Journal of Biomedical and Health Informatics 2018 42 assisted The paper's contribution is a human-designed classification method (multimodal DPN plus regularized extreme learning machine) whose performance is benchmarked on existing public PPI datasets, so the AI's outputs are accuracy evaluations on known data rather than an end-to-end AI-produced scientific result. “The proposed method is tested on the public dataset of H. pylori, Human, and Yeast and achieves average accuracies of 97.87%, 99.90%, and 98.11%, resp...”
APOGEE 2: multi-layer machine-learning model for the interpretable predictio... Nature Communications 2023 42 assisted APOGEE 2 is a pathogenicity-scoring tool whose precomputed predictions serve as one input into the human-run ACMG/AMP clinical interpretation workflow, rather than an AI-produced result that itself constitutes the paper's primary outcome. “With clinical and research use in mind, APOGEE 2 scores and pathogenicity probabilities are precompiled and available in MitImpact.”
A machine learning approach for wind turbine power forecasting for maintenan... Energy Informatics 2025 41 assisted The paper's contribution is a comparison of nine existing ML models' forecasting accuracy on an existing dataset, with the forecasts serving as input to human-driven maintenance planning rather than the AI performing the task end to end. “This study presents and compares nine machine learning (ML) methods for forecasting... Computational results show that the GNN-based forecasting model...”
Therapeutic enzyme engineering using a generative neural network Scientific Reports 2022 40 assisted The generative model is one component of a combined computational-and-wet-lab engineering pipeline: it designs candidate sequences that are then constructed and experimentally validated, so the AI informed rather than itself produced the improved enzyme variants. “combines deep latent variable modelling of sequence co-evolution with automated protein library design and construction to rapidly identify metabolic...”
CRISPRLearner: A Deep Learning-Based System to Predict CRISPR/Cas9 sgRNA On-... Electronics 2019 39 assisted The system only predicts sgRNA cleavage efficiency as a support tool within a human-driven design workflow, with its contribution being a correlation metric on existing datasets rather than an end-to-end accomplishment. “CRISPRLearner supports researchers in sgRNA design task, predicting a sgRNA on-target knockout efficiency.”
The impact of chemotherapy and survival prediction by machine learning in ea... BMC Geriatrics 2022 37 assisted The machine learning models are survival-prediction components within a human-run epidemiological workflow (propensity matching, multivariable survival analysis) whose output informs chemotherapy treatment decisions rather than the AI itself producing the clinical result. “The LightGBM is a trustful model to predict the survival and provide precious systemic treatment for patients received chemotherapy.”
CLECG: A Novel Contrastive Learning Framework for Electrocardiogram Arrhythm... IEEE Signal Processing Letters 2021 35 assisted This is a methods paper whose contribution is a human-designed contrastive pre-training framework, evaluated by reporting comparative accuracy on the existing task of ECG arrhythmia classification, rather than the AI itself producing a novel solved result. “Experimental results show that the proposed CLECG strategy outperforms other self-supervised methods and supervised transfer learning strategies.”
Detection of sarcopenic obesity and prediction of long‐term survival in pati... Journal of Surgical Oncology 2021 34 assisted The paper's contribution is a clinical assessment of prevalence and prognostic value of sarcopenic obesity in gastric cancer patients, with machine learning serving as a component tool within a human-run prognostic study rather than being the end-to-end solved task. “we assessed the prevalence and prognostic value of sarcopenic obesity in a large population of gastric cancer patients using preoperative CT”
Large Language Model Assistant for Emergency Department Discharge Documentat... JAMA Network Open 2025 33 assisted The LLM only drafts discharge notes that emergency physicians still review and finalize, making it a component in a human-run documentation workflow rather than the agent producing the end result. “the LLM-assisted notes were more complete, correct, concise, and clinically useful than manual notes”
Microbiome meta-analysis and cross-disease comparison enabled by the SIAMCAT... bioRxiv (Cold Spring Harbor Laboratory) 2020 32 assisted The paper's primary contribution is a software toolbox, and the ML models within it served as components of a human-designed meta-analysis workflow whose biomarker findings were interpreted by the researchers, not produced end-to-end by an AI agent. “we developed SIAMCAT, a versatile R toolbox for ML-based comparative metagenomics”