Research Applications

Evaluating the effectiveness of general-purpose Tabular Foundation Models such as TabPFN against specialized models in predicting cellular responses to genetic and chemical perturbations across various biological datasets and scales.

Optimizing catalyst formulation and operating parameters for the water–gas shift reaction to improve CO conversion, using an interpretable TabPFN-ensemble framework.

This study uses features extracted from Transcranial Doppler (TCD) to estimate chronological brain vascular age and assess accelerated aging in subjects with various brain diseases, utilizing machine learning regression models trained on healthy subjects.

An interpretable machine learning framework using TabPFN predicts reaction rate constants for antibiotic degradation catalyzed by biochar. The model outperforms other ML techniques with high accuracy and interpretability, revealing key catalyst properties and…

This study develops a machine learning ecosystem to improve cardiovascular disease diagnosis through early warning systems using self-reported features and specialized diagnostic models integrating clinical and non-clinical data. It uses advanced ML methods i…

Predicting recovery outcomes after spinal cord injury using data-driven modeling, with a focus on motor status and walking function prediction based on early clinical assessments from the Sygen trial dataset.

The study developed a transcriptome-based approach combining bulk RNA-seq data from kidney cell lines, primary cells, or tissues with single-cell RNA-seq reference datasets to accurately match and validate kidney cell types. TabPFN, alongside Spearman correla…

A method for survival analysis in medical applications using a prior-fitted in-context learning model pretrained on synthetic data to enable individualized survival prediction without task-specific training.

This study presents a fully automated methodology for early prediction studies in clinical settings, leveraging information extracted from unstructured discharge reports. The proposed pipeline supports cohort selection, dataset generation, and outcome labelin…

This study uses the TabPFN model to predict the International Roughness Index (IRI) of asphalt pavements by integrating traffic load, pavement structure parameters, and climate variables, outperforming traditional regression models and other machine learning…

This use case addresses water inrush risks in coal mining by integrating 2D discrete-element simulation (PFC2D) with TabPFN for small sample prediction and SHAP for explainability, enabling accurate and interpretable prediction of water-conducting fracture zo…

TabPFN is applied to early-stage drug discovery where data scarcity and out-of-distribution shifts limit the reliability of predictive models. It offers robust and data-efficient tabular modeling, outperforming traditional methods like XGBoost in regression t…

Application of TabPFN and its tabpfn-extensions library to geotechnical tasks including soil-type classification and iterative imputation of mechanical parameters, with visualization and uncertainty quantification aiding interpretability in data-scarce scenar…

Development of a delay-aware digital-twin MPC benchmarking framework for anaerobic digesters using various surrogate ML models including TabPFN to provide fast one-step predictions under irregular measurements, enabling improved methane production and robust…

Development of a machine learning framework for postoperative breast cancer related lymphedema (BCRL) risk prediction using routinely collected clinical data from breast cancer patients. Various classifiers, including transformer-based TabPFN, were evaluated…

This use case involves using TabPFN as a surrogate model in active learning for materials discovery. TabPFN replaces conventional Gaussian Process and Random Forest models to provide better uncertainty calibration and regression performance, enabling more dat…

This study proposes a symmetric leaf-vein (SLV) texture to improve thermohydrodynamic lubrication performance of journal bearings. It uses a surrogate-assisted multi-objective optimization combining a TabPFN surrogate model with NSGA-II to optimize texture pa…

This use case involves predicting the maximum loading of alkane isomers in zeolites for hydroisomerization processes using a cascade machine learning model including a TabPFN pretrained model. The model improves prediction accuracy of adsorption capacity thro…

Development of an accurate, low-cost, and clinically feasible early screening model to identify women at high risk of recurrent spontaneous abortion (RSA) using routinely available clinical biomarkers, leveraging a Transformer-based tabular model (TabPFN) alo…

Using TRAMs derived from delayed-contrast MRI and radiomic features to distinguish malignant breast lesions from benign ones, improving diagnostic precision and potentially reducing unnecessary biopsies in breast cancer diagnosis.

Using a multimodal multi-instance learning framework to predict peak oxygen consumption (peak VO₂) from transthoracic echocardiography studies and electronic health records, improving accuracy and generalization for heart failure patient assessment.

This study develops a machine learning framework using TabPFN to assess slope stability based on geometric features and shear strength parameters. It integrates field investigations with numerical simulations to create a 2D slope database. TabPFN outperforms…

This use case involves the development and application of transformer-based probabilistic deep learning models to predict the success of hematopoietic stem cell (HSC) mobilization in healthy donors using baseline and post-mobilization laboratory data. The mod…

This use case focuses on predicting the fracture energy (Gc) and Mode I fracture toughness (KIC) of concrete structures using 25 machine learning algorithms including classical models and modern transformer-inspired neural networks like TabPFN. TabPFN outperf…

This use case addresses the classification of adolescent alcohol drinking behavior using low-cost, accessible clinical measures rather than expensive neuroimaging data. The proposed framework, FocalTab, integrates TabPFN with focal loss to improve accuracy an…

Development of ¹H NMR-based metabolomics-AI platforms integrating serum metabolomic profiles and clinical parameters to enable rapid, accurate, and non-invasive early detection of pancreatic ductal adenocarcinoma (PDAC).

Using TabPFN to approximate conditional expectations for Shapley value computation in explainable AI, achieving better or comparable performance to state-of-the-art methods with lower runtime.

Machine learning models were trained to predict adverse events requiring hospitalization and forecast AE onset timing for optimized ward management in myeloma patients undergoing stem-cell mobilization and collection, aiming to enable a risk-stratified outpat…

ExplainerPFN is a tabular foundation model built on TabPFN, pretrained on synthetic structural causal datasets supervised with exact or near-exact Shapley values, designed to predict feature attributions for unseen tabular datasets without model access, gradi…

This work proposes a pre-trained model based framework for predicting cellular toxicity of silica nanoparticles. It addresses data leakage by removing evaluation stage features and uses the embedding layer from TabPFN to convert categorical values into dense…

This study uses AI models including TabPFN to predict and optimize the compressive strength of fly ash-based geopolymer concrete, achieving high accuracy and stability in predictions and combining metaheuristic algorithms for mix design optimization.

Using a descriptor-first strategy combined with a pretrained foundation model for tabular data (TabPFN) for single-task regression in predicting ADMET properties relevant to small-molecule drug design, achieving improved accuracy and simplified deployment com…

Study on using early arterial lactate trajectory patterns to predict mortality risk in ICU-admitted heart failure patients, identifying distinct lactate classes that correlate with different mortality outcomes and providing enhanced prognostic insights for ea…

This study evaluates the use of TabPFN for modeling volatile organic compound (VOC) profiles obtained via mass spectrometry methods like gas chromatography–mass spectrometry and proton transfer reaction mass spectrometry. TabPFN is applied to high dimensional…

This study benchmarks nine open-source AutoML frameworks and the transformer-based model TabPFN across 25 financial predictive analytics datasets, evaluating their performance in various financial tasks. It finds that TabPFN excels in multiclass tasks within…

This use case involves applying various machine learning techniques to predict the compressive strength of geopolymer concrete, using a dataset of 204 instances including experimental data. Features such as chemical composition and curing temperature are used…

This study evaluates how measurement errors in respiratory rate and heart rate affect the accuracy of predicting the success or failure of high flow nasal cannula (HFNC) therapy in patients with acute hypoxemic respiratory failure. It compares the TabPFN mach…

The study uses the Optuna-TabPFN model to accurately predict the unconfined compressive strength (UCS) of cemented paste backfill (CPB) with limited data, achieving high accuracy and employing SHAP analysis to interpret the model and identify key influencing…

This study introduces a bias-aware and accurate predictive framework integrating TabPFN with Monte Carlo Bias Estimation (MCBE) to model the elastic modulus of recycled aggregate concrete (RAC), enabling reliable, bias-corrected, and interpretable predictions…

A deep-learning surrogate model based on TabPFN predicts free energy with high fidelity to assess synthetic feasibility of hypothetical metal-organic frameworks (hMOFs), enabling rapid flagging of prime MOF candidates for synthesis and accelerating free-energ…

This research applies machine learning models, including TabPFN, to predict crack mouth opening displacement (CMOD) in fiber-reinforced ultra-high-performance concrete (FR-UHPC). The study uses a comprehensive experimental database and compares several ML alg…

PSF-Net combines TabPFN’s meta-learned prior knowledge and SAINT’s dual attention mechanism with an uncertainty-aware gating block to predict downlink power density and classify radiation levels from 5G base stations. The model provides calibrated uncertainty…

This use case involves using TabPFN, a pre-trained transformer model for tabular data classification, to differentiate between benign and malignant renal tumors based on CT radiomics datasets. Compared with conventional machine learning algorithms, TabPFN sho…

The study uses artificial intelligence frameworks, including the TabPFN model, to distinguish and classify corneal stromal and endothelial disorders from controls and to categorize corneas into seven groups for diagnosis with high accuracy.

Using machine learning models augmented with TabPFN data augmentation to predict capacity and initial Coulombic efficiency of biomass-derived hard carbon for sodium-ion batteries, identifying key synthesis parameters and improving screening efficiency.

This use case focuses on predicting ship fuel consumption using various modeling approaches, including tabular and time-series regression models. It benchmarks these models on a new dataset comprising operational and environmental data from three ships and in…

A novel foundation model-based recommendation framework using TabPFN to integrate multi-omics profiles with clinical factors for personalized neoadjuvant therapy recommendations in breast cancer, improving treatment response rates significantly.

This study developed and clinically validated an interpretable Transformer-based predictive model, TabPFN-V2, to assess gastrointestinal bleeding risk in pediatric patients with Henoch-Schönlein purpura (IgA vasculitis). The model used comprehensive clinical…

This study investigates an uncertainty-guided strategy for model selection using TabPFN models in biomolecule efficacy prediction, specifically on an siRNA knockdown efficacy task. The proposed OligoICP method selects an ensemble of TabPFN models with the low…

This study evaluated the predictive performance of various machine learning and deep learning models including TabPFN in estimating manual strength in men and women using anthropometric data collected from 382 participants. TabPFN and TabNet deep learning mod…

Development of a machine learning framework to rapidly and accurately predict Henry coefficients of linear and branched alkanes in one-dimensional zeolites, enabling efficient screening and design of shape-selective zeolite catalysts to support catalytic hydr…

This use case focuses on applying small-data machine learning models such as TabPFN for forecasting individual-level mental health trajectories using personalized digital phenotyping data. It addresses challenges of sparse, irregular, and noisy data in contin…

Development and validation of an interpretable transformer-based clinical prediction model (TabPFN-V2) for identifying IVIG resistance in pediatric Kawasaki disease patients, demonstrating superior predictive performance and interpretability to support person…

The study applies the TabPFN algorithm to combine delta-radiomics features derived from dynamic MRI with clinical and morphological features to improve the preoperative prediction of lymphovascular invasion in invasive breast cancer patients.

Development of non-invasive predictive models for diagnosing minimal change disease (MCD) using machine learning algorithms including TabPFN. The study compares TabPFN’s performance against other models and identifies key predictors using SHAP.