Research
-
Resource Inversion: Cross-Lingual Fidelity and Verified Deferral for Oncology Patient Communication in Low-Resource Languages
-
A 6-Gene pSTAT3 Transcriptomic Score Identifies an Immunosuppressive, Chemotherapy-Resistant Phenotype and Predicts Poor Survival in Biliary Tract CancerDevelops and validates a six-gene pSTAT3 transcriptomic score in biliary-tract cancer, linking high pathway activity to shorter survival, an immunosuppressive tumor microenvironment, and chemotherapy resistance.
-
Tissue tearing degrades optimal-transport registration of spatial transcriptomics beyond displacement magnitude: a multiseed deformation benchmark and a supervised graph cross-attention proof-of-conceptBuilds a controlled human-DLPFC benchmark showing that tissue tears break optimal-transport, diffeomorphic, and Gaussian-process registration methods; tests a graph cross-attention model and diagnoses donor shift as its main generalization failure.
-
Napabucasin Across Four Registrational Phase 3 Trials in Advanced Gastrointestinal Cancers: A Complete-Enumeration Meta-Analysis Bounding the Survival Effect, and a Tested Non-Replication of the pSTAT3 Biomarker SignalPools all four registrational phase 3 trials of napabucasin, covering 3,383 patients, to bound any clinically meaningful survival benefit and test whether the reported pSTAT3-positive biomarker signal replicates.
-
A pan-cancer identifiability atlas for compartment-resolved pathway attribution from bulk transcriptomesIntroduces a calibrated four-part identifiability gate for asking whether bulk RNA can recover malignant-versus-stromal pathway differences; across 5,576 TCGA tumors, only 6.9% of pathway-cohort attributions pass.
-
Competing risks are ignored in most SEER cancer-specific-mortality nomograms: a systematic audit with real-patient reanalysis and a validated correction pipelineAudits 1,000 SEER nomogram papers and finds that 82.8% show no evidence of competing-risk handling; validates how naive survival methods overestimate cancer-specific mortality and releases a reproducible correction pipeline.
-
Most published immunotherapy-response gene signatures do not outperform random gene sets, and collapse onto a single tumor-inflammation axisBenchmarks 102 published immune-checkpoint response signatures across 15 cohorts against size-matched random gene sets, finding that most collapse onto one tumor-inflammation axis with a held-out predictive ceiling near AUROC 0.59.
-
Tumor Phospho-STAT3 Marks a Shared Angiogenic-Fibrotic, Immune-Replete Microenvironment Across Gastrointestinal Cancers: A Multi-Omic Analysis of 1,274 TumorsUses genomic, transcriptomic, and spatial data from 1,274 tumors across gastrointestinal cancers to test whether pSTAT3 marks a shared angiogenic-fibrotic, immune-replete microenvironment linked to patient survival.
-
When can a pathway program be attributed to a tumour compartment from bulk RNA? An identifiability audit across immune-checkpoint cohortsAudits whether malignant and stromal STAT3 activity can be separated from bulk RNA-seq in six immune-checkpoint cohorts; only two pass the identifiability gate, and compartment resolution adds no predictive value beyond the bulk signal.