XGBoost is a gradient-boosted decision tree method used as a strong baseline comparator for tabular machine learning, combining many weak trees into an ensemble to improve predictive accuracy. In the provided studies, it was applied across classification and regression tasks, including immunotherapy response prediction, hepatocellular carcinoma diagnosis, antibiotic resistance modeling, and half-life prediction for biologics and ADCs. A key recent finding is that tabpfn matched XGBoost in classification and exceeded it in regression in many settings, reinforcing XGBoost’s role as a benchmark for small-data tabular learning. In pan-cancer modeling, XGBoost used 14 high-performance regulatory features to predict immunotherapy response and outperformed traditional markers such as pd l1 in distinguishing responders from non-responders. It also achieved the best predictive performance for a hepatocellular carcinoma diagnostic signature and was used to train multimodal half-life models for antibody drug conjugates and monoclonal antibodies.
Cancer
- A pan-cancer XGBoost model integrated 14 regulatory features to predict immunotherapy response and outperformed traditional markers such as pd l1 in separating responders from non-responders. (PMID:41925746)
- XGBoost was used as the best-performing classifier for a hepatocellular carcinoma diagnostic gene signature. (PMID:41671946)
- The hepatocellular carcinoma study combined transcriptomics, machine learning, network pharmacology, and molecular dynamics to support biomarker discovery and drug repurposing. (PMID:41671946)
- In tabular drug-discovery benchmarking, TabPFN matched XGBoost in classification and surpassed it in regression in many settings, underscoring XGBoost as a dominant baseline. (PMID:41867095)
Delivery / Pharmacokinetics
- XGBoost models were trained on multimodal features to prioritize antibody drug conjugates with favorable half-life in mice. (PMID:41996252)
- The same multimodal framework was used to train half-life prediction models for monoclonal antibodies. (PMID:41996252)
- These results highlight XGBoost’s utility for pharmacokinetic property prediction from heterogeneous feature sets. (PMID:41996252)
Microbiology / Resistance Prediction
- XGBoost was among the top-performing ensemble methods evaluated on MALDI-TOF-MS spectra for predicting antibiotic resistance in Klebsiella pneumoniae. (PMID:41771780)
- The study supports gradient boosting as a competitive approach for spectral classification in clinical microbiology. (PMID:41771780)