My research moves from molecular epidemiology of glioma through computational methods to clinical AI translation — connecting biological mechanism, patient outcome, and the systems that make rigorous research scalable.
Experience of a 3-year pilot at UCSF's Brain Cancer Survivorship Program which trained 46 peer volunteers to support fellow brain tumor patients, completing 132 peer matches with high volunteer satisfaction and sustained engagement — demonstrating that structured, ongoing group support can effectively sustain patient peer volunteer programs in neuro-oncology.
Genomic profiling of 205 IDH-mutant astrocytomas revealed that RAS-MAPK pathway alterations are enriched in recurrent, high-grade gliomas and associated with inferior survival, implicating this pathway as a key driver of malignant transformation and a potential therapeutic target.
A systematic review of 11 randomized controlled trials found that while neuro-oncology caregiver outcome reporting generally meets quality standards, critical methodological gaps — particularly around missing data handling, participant flow, and data collection windows — warrant standardization efforts to strengthen the evidence base for caregiver support interventions.
GPT-4 and Claude-2 were both capable of generating clinically valid molecular subtyping decision trees across five cancer types, though adding expert-curated guideline text offered only modest accuracy gains — suggesting LLMs hold real promise for distilling oncologic diagnostic frameworks but require careful oversight before clinical deployment.
Among 1,504 glioma patients spanning pre- and post-TMZ treatment eras, germline variants in DNA repair genes — most notably an MGMT SNP and elevated PNKP expression linked to a significant survival difference in IDH-mutant tumors — were associated with TMZ sensitivity in a subtype-specific manner, pointing toward germline-informed chemotherapy personalization as a tractable next step.
