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
An effective self-supervised framework for learning expressive molecular glo... Briefings in Bioinformatics 2021 151 assisted The paper contributes a pre-training framework/encoder whose value is demonstrated as accuracy on 14 existing benchmark prediction tasks, positioning the AI as a molecular-encoder component for human-run drug discovery pipelines rather than as the agent that solved a discovery problem. “The pre-trained MolGNet in MPG has the potential to become an advanced molecular encoder in the drug discovery pipeline.”
A Machine Learning Approach for the Differential Diagnosis of Alzheimer and... Frontiers in Neuroinformatics 2020 133 assisted The paper reports the classification accuracies of ML algorithms (ANN, SVM, ANFIS) on an existing diagnostic task, framing the system as a decision-support tool for physicians rather than an agent that performs the differential diagnosis end to end. “therefore suggesting its usefulness in supporting physicians' diagnostic evaluations”
Expansion of RiPP biosynthetic space through integration of pan-genomics and... PLoS Biology 2020 127 assisted The SVM-based decRiPPter only prioritized candidate RiPP gene clusters, while the actual discovery, elucidation of the novel class V lanthipeptide pristinin A3, was completed by humans through NMR, MS/MS, and chemical labeling. “The 2D structure of the new RiPP, which we name pristinin A3 (1), was solved using nuclear magnetic resonance (NMR), tandem mass spectrometry (MS/MS)...”
Damage detection in a novel deep-learning framework: a robust method for fea... Structural Health Monitoring 2019 110 assisted The paper's contribution is a new CNN architecture whose performance is reported as accuracy gains on existing damage-detection benchmark datasets (numerical simulations and lab measurements), i.e., an improved predictive tool within the human-run structural health monitoring workflow rather than the AI itself producing a novel solved result. “The results show that our proposed method has accuracy improvements of at least 10% over other network architectures.”
Chest X-Ray Analysis of Tuberculosis by Deep Learning with Segmentation and... IEEE International Conference on Electronics and Nanotechnology (ELNANO) 2018 109 assisted The paper reports a CNN's validation loss/accuracy on an existing TB-detection task, with its actual contribution being the methodological comparison of segmentation and augmentation strategies rather than the AI end-to-end solving the diagnostic problem. “They demonstrate the efficiency of lung segmentation, lossless and lossy data augmentation for CADx of tuberculosis by deep convolutional neural netwo...”
Fitting potential energy surfaces with fundamental invariant neural network.... The Journal of Chemical Physics 2020 103 assisted The paper's contribution is a human-developed computational scheme for generating fundamental invariants (based on invariant theory and parallel programming) that serves as input to neural networks, making the NN merely one fitting component inside a human-designed PES construction workflow rather than the agent that produced the result. “we report an improved and efficient generation scheme of FIs based on the computational invariant theory and parallel program, which can be readily us...”
Generative deep learning enables the discovery of a potent and selective RIP... Nature Communications 2022 101 assisted The GDL model generated a virtual candidate library, but the actual discovery of the validated RIPK1 inhibitor came from a human-run pipeline of virtual screening, synthesis, and in vitro/in vivo bioactivity evaluation, making the AI one component in a workflow humans completed. “Virtual screening against the generated tailor-made compound library and subsequent bioactivity evaluation lead to the discovery of a potent and selec...”
A Deep Learning Model for Cervical Cancer Screening on Liquid-Based Cytology... Cancers 2022 91 assisted The deep learning model reports classification performance (ROC AUCs) on an existing screening task, but the abstract explicitly frames it as a pilot study with potential for 'aiding' human-run screening processes rather than the AI itself performing cervical cancer screening end to end. “demonstrating the promising potential use of such models for aiding the screening processes for cervical cancer”
Single-cell multiome of the human retina and deep learning nominate causal v... Cell Genomics 2022 88 assisted Deep learning models are one predictive component within a human-driven multi-omics pipeline (single-cell atlas, HiChIP, eQTL integration), and the outcome is nomination of candidate variant-gene pairs rather than the AI itself producing a solved result. “Our efforts nominate pathogenic SNP-target gene interactions for multiple vision disorders”
LM-GVP: an extensible sequence and structure informed deep learning framewor... Scientific Reports 2022 85 assisted This is a methods paper whose reported results are benchmark accuracy on existing property-prediction tasks, with the framework explicitly positioned as a tool to accelerate future protein engineering and drug development rather than an AI-produced solution to a scientific problem. “Our approach outperformed the state-of-the-art protein LMs on a variety of property prediction tasks including fluorescence, protease stability, and p...”
Evaluation of an artificial intelligence clinical trial matching system in A... JAMIA Open 2020 85 assisted The paper evaluates an AI trial-matching tool whose eligibility predictions are prescreening outputs within a clinician-led workflow, with the gold standard and final decisions made by clinicians, making it a decision-support component rather than an end-to-end solution. “however, clinician input and oversight are still required. The automated system demonstrates promise as a clinical decision support tool to prescreen...”
Machine learning-guided co-optimization of fitness and diversity facilitates... Nature Communications 2024 83 assisted MODIFY's ML predictions and library designs guided a human-run wet-lab directed evolution workflow to produce the biocatalysts, rather than the AI itself performing the enzyme engineering end to end. “Using MODIFY, we engineer generalist biocatalysts derived from a thermostable cytochrome c to achieve enantioselective C-B and C-Si bond formation”
Automated Explainable Multidimensional Deep Learning Platform of Retinal Ima... JAMA Network Open 2021 83 assisted This is a diagnostic accuracy study benchmarking a deep learning platform's classification performance against expert ophthalmologists on an existing clinical task, with the platform positioned as a decision-support aid for routine human-run ROP screening rather than an autonomous solver. “Results suggest that a deep learning platform could identify and classify multidimensional ROP pathological lesions in retinal images with high accura...”
DeepPVP: phenotype-based prioritization of causative variants using deep lea... BMC Bioinformatics 2019 78 assisted The abstract reports a deep-learning tool's comparative benchmark accuracy on the existing prediction task of variant prioritization, where the ranked 'likely causative' variants are candidate outputs that inform human geneticists in a diagnostic workflow rather than an AI-produced end result. “We demonstrate that DeepPVP performs significantly better than existing methods, including phenotype-based methods that use similar features.”
Assessment of deep learning assistance for the pathological diagnosis of gas... Modern Pathology 2022 76 assisted The study evaluates deep learning as a decision-support tool for human pathologists who made the diagnoses themselves, with the AI only informing their reads rather than producing the diagnostic result. “Our results demonstrated that DL assistance indeed improved pathologists' accuracy and efficiency in gastric cancer diagnosis”
DeepWAS: Multivariate genotype-phenotype associations by directly integratin... PLoS Computational Biology 2020 74 assisted The deep learning component only supplies per-variant regulatory effect predictions that are fed as features into a human-designed multivariate GWAS pipeline, so the AI is one input component rather than the agent producing the disease associations. “We here describe "DeepWAS", a new approach that integrates these regulatory effect predictions of single variants into a multivariate GWAS setting.”
Crystal twins: self-supervised learning for crystalline material property pr... npj Computational Materials 2022 72 assisted The paper's contribution is a human-designed self-supervised method whose value is demonstrated as improved prediction accuracy on existing benchmarks, making the GNN a prediction component in an established evaluation task rather than the agent that solved a materials problem. “we significantly improve the performance of GNN on 14 challenging material property prediction benchmarks”
Deep learning enhancing guide RNA design for CRISPR/Cas12a‐based diagnostics iMeta 2024 63 assisted The deep learning model predicts and ranks crRNA candidates, but the effective diagnostic reagents are established through human-run wet-lab CRISPR assays, making the AI a prediction component inside a human-completed assay-design and validation workflow rather than the agent that produced the end result. “all the top five predicted crRNAs for each HPV subtype exhibited robust fluorescent signals in CRISPR assays, thereby suggesting that the platform cou...”
Convolution Neural Networks and Self-Attention Learners for Alzheimer Dement... Sensors 2023 62 assisted The paper is a human-designed and human-run comparative benchmark of existing off-the-shelf architectures on public datasets, where the researchers evaluate model accuracies rather than the AI itself producing the paper's primary result. “The contribution of this research was to investigate for the first time, to our knowledge, three current and promising deep convolutional models (ResN...”
Determination of disease severity in COVID-19 patients using deep learning i... Diagnostic and Interventional Radiology 2020 61 assisted The primary results (clinical associations with opacity severity) came from human radiologist review, and the deep learning model's role was limited to demonstrating feasibility of opacity detection, validated against human interpretation with only moderate concordance (kappa 0.51-0.71), rather than itself producing the severity determinations that constitute the paper's contribution. “In comparison to human interpretation, the deep learning model achieved a kappa concordance of 0.51 for alveolar opacities and 0.71 for interstitial o...”
Lung Cancer Survival Prediction via Machine Learning Regression, Classificat... IEEE International Symposium on Signal Processing and Information Technology (ISSPIT) 2018 61 assisted The paper's primary contribution is a human-designed evaluation reporting accuracy metrics of standard off-the-shelf models (Random Forests, GBM, linear regression) on the existing task of SEER-based survival prediction, rather than an AI system whose own predictive output constitutes the solved result. “The regression results show that RF had the best performance for survival times ≤6 and >24 months (RMSE 10.52 and 20.51, respectively), while GBM perf...”
Application of machine learning and genetic optimization algorithms for mode... PLoS ONE 2021 59 assisted The ML/GA system produced a yield-prediction model and a suggested ideotype, but the abstract frames these outputs as decision support for human breeders who will actually select parents and make crosses, making the AI a component in a human-run breeding pipeline rather than the agent that solved the problem. “which can be used for selecting parental lines and designing promising crosses for developing cultivars with improved genetic yield potential”
Application of Machine Learning Models for Survival Prognosis in Breast Canc... Information 2019 56 assisted The paper applies standard existing ML regression models to report prediction accuracy on breast cancer survival data within a human-run clinical workflow intended to inform therapy and management decisions, rather than the AI itself producing the end result as the paper's core contribution. “the applied machine learning methods provide promising results in terms of accuracy of survival time prediction”
Deep Learning for Concrete Crack Detection and Measurement Metrology 2024 55 assisted The deep learning models serve as classification/segmentation components within a human-engineered three-stage pipeline, and the paper's novel contribution (millimetre crack measurement) comes from the non-AI laser calibration method, with accuracies reported on existing crack detection tasks. “The final stage involved measuring crack width in millimetres by using a novel laser calibration method.”
Machine learning-based prediction of survival prognosis in cervical cancer BMC Bioinformatics 2021 53 assisted Humans designed and ran the entire TCGA analysis pipeline (preprocessing, Cox feature selection, K-means clustering, model construction) to produce a prognostic stratification tool whose survival predictions exist to inform clinical management, so the ML was the analytical method in a human-driven study rather than an agent that itself produced a standalone result. “The aim of this study is to improve forecasting capability by developing a miRNAs-based machine learning survival prediction model.”
The Use of Machine Learning Methodologies to Analyse Antibiotic and Biocide... PLoS ONE 2013 52 assisted Machine learning was used as an exploratory analytical tool inside a human-designed and human-executed study (researchers collected 1632 isolates, performed all phenotypic MIC characterization, and drew the scientific conclusions from the ML outputs), rather than an AI system itself producing the paper's result end to end. “Therefore, machine learning techniques were applied to explore the data.”
Interpretable machine learning model for early prediction of 28-day mortalit... European journal of medical research 2024 52 assisted This is a development-and-validation study of a clinical risk prediction model whose output is intended to inform clinicians' decisions rather than the AI itself performing or completing the clinical task end to end. “The model established will have the potential to improve the level of clinical practice”
CONNECT: a neural network based framework for emulating cosmological observa... Journal of Cosmology and Astroparticle Physics 2023 49 assisted The neural network is an emulation component (a plug-in) inside a human-orchestrated MontePython inference pipeline, humans generate the CLASS training data, design the training, and run the sampling, so the AI accelerates a workflow humans still run rather than producing the scientific result end-to-end. “a neural network framework emulating class computations as an easy-to-use plug-in for the popular sampler MontePython”
Unlocking HDR-mediated nucleotide editing by identifying high-efficiency tar... Scientific Reports 2019 48 assisted The machine learning model predicts high-efficiency target sites and ssODN designs to guide researchers, but the actual HDR-mediated nucleotide editing is performed by humans in wet-lab experiments, making the AI a design-informing component of a human-run pipeline. “we developed CUNE (Computational Universal Nucleotide Editor), which enables users to identify and design the optimal targeting strategy”
Polymer design via SHAP and Bayesian machine learning optimizes pDNA and CRI... Chemical Science 2024 47 assisted The ML (SHAP interpretation and Bayesian optimization) is one component in a human-run wet-lab workflow of polymer synthesis, characterization, and in vivo validation, informing rather than itself producing the delivery result. “Our facile coupling of synthesis, characterization, and machine analysis provides powerful tools to quantitate performance parameters accelerating nex...”
Detecting cardiomyopathies in pregnancy and the postpartum period with an el... European Heart Journal - Digital Health 2021 47 assisted The paper reports retrospective accuracy metrics (AUC, sensitivity, specificity) for a deep learning model positioned as a screening aid whose outputs would inform clinicians, with definitive diagnosis grounded in echocardiography, a model evaluation inside a human clinical pipeline rather than the AI itself solving the diagnostic task. “with the potential to become a powerful initial screening tool for cardiomyopathy in the obstetric care setting”
Deep structured learning for variant prioritization in Mendelian diseases Nature Communications 2023 43 assisted MAVERICK is a pathogenicity-prediction tool benchmarked on already-solved patient cohorts, and its real-world impact is described as 'facilitating' human identification of a novel disease gene, meaning its predictions informed humans who completed the diagnostic/gene-discovery workflow rather than the AI itself producing the discovery. “MAVERICK has already facilitated the identification of a novel disease gene causing a degenerative motor neuron disease.”
Learning deep representations of enzyme thermal adaptation Protein Science 2022 42 assisted The paper's contribution is a prediction model benchmarked on existing tasks (predicting optimal catalytic and melting temperatures), with the actual engineering of thermostable enzymes left as future human work the model merely 'guides'. “DeepET thus holds promise for understanding enzyme thermal adaptation and guiding the engineering of thermostable enzymes.”
Optimal tuning of three deep learning methods with signal processing and ano... Structural Health Monitoring 2024 42 assisted The deep learning models serve as classifiers inside a human-engineered pipeline (MEMD/Wavelet feature extraction, meta-heuristic tuning), and the contribution is reported classification accuracies on bridge monitoring data rather than the AI autonomously performing damage assessment end to end. “The average accuracy values are obtained as 98.13, 97.96, and 97.79 for 1D-CNN, LSTM, and MLP, respectively.”
Determining a minimum set of variables for machine learning cardiovascular e... Cardiovascular Research 2021 42 assisted The paper reports ML model accuracy on an existing task (MACE risk stratification from SPECT MPI) with the actual contribution being a human-driven feature-reduction analysis to make the model practical for physicians, so the AI is a component informing a clinical workflow rather than the agent whose output constitutes the contribution. “ML models with fewer collected variables may be more practical for clinical implementation.”
Formation of a Lightweight, Deep Learning-Based Weed Detection System for a... Applied Sciences 2023 41 assisted The AI is a weed-detection component inside a human-built laser weeding robot pipeline, and the abstract only reports benchmark metrics (mAP, FPS) plus a calculated field-time estimate rather than the AI itself performing the weeding task end to end. “This work focuses on developing a lightweight weed detection mechanism to assist laser weeding robots.”
Uncertainty estimation for deep learning-based automated analysis of 12-lead... European Heart Journal - Digital Health 2021 41 assisted The paper's contribution is a methodological evaluation of uncertainty-estimation techniques around a DNN ECG classifier within a human-in-the-loop workflow where uncertain cases are deferred to physicians, not the AI end-to-end solving the diagnostic task. “The clinical simulation showed that the accuracy of the algorithm increased as uncertain predictions were referred to the physician.”
A Deep-Learning View of Chemical Space Designed to Facilitate Drug Discovery Journal of Chemical Information and Modeling 2020 41 assisted DESMILES is a generative design tool whose reported results are failure-rate metrics on a previously published benchmark and computational docking scores; it performs only the design step of the human-run drug-discovery cycle with no experimentally validated discovery, so the AI augments rather than solves the problem. “The use of deep learning technologies could augment the typical practice of using human intuition in the design cycle, and thereby expedite drug disco...”
Benchmark analysis of day-ahead solar power forecasting techniques using wea... Conference record of the IEEE Photovoltaic Specialists Conference 2019 40 assisted The paper is a benchmark evaluation reporting accuracy/skill scores of existing ML techniques on the established task of day-ahead PV forecasting, with the contribution being the comparative analysis rather than a new end-to-end capability. “With a skill score of respectively 41.4% and 41.3% Gradient Boosting and Random Forests are found to outperform the other models.”
Learning to predict RNA sequence expressions from whole slide images with ap... Communications Biology 2023 40 assisted The paper proposes a model whose RNA-seq predictions are benchmarked against state-of-the-art and explicitly framed as a tool to assist downstream computational pathology workflows (search and classification), not as an AI-produced result that itself constitutes the paper's finding. “The proposed tRNAsformer can assist as a computational pathology tool to facilitate a new generation of search and classification methods by combining...”