Selected research
Six projects, one throughline.
Computational and translational oncology across lung, breast, and gynecologic cancers — most in active preparation for publication and presentation at AACR.
Nov 2025 — presentIndependent · Emory University mentorshipLead researcher
Computational drug repurposing in NSCLC: targeting the EGFR L858R oncogenic mutation
Non-small-cell lung cancer accounts for roughly 85% of lung-cancer diagnoses, and the L858R substitution in exon 21 of EGFR locks the kinase in an active conformation, driving constitutive RAS/MAPK and PI3K/AKT/mTOR signaling. Rather than chase a novel hit on a decade-plus timeline, this screen repurposes FDA-approved non-oncology compounds — molecules that already carry established human safety and pharmacokinetic profiles — for meaningful binding at the L858R site. Candidate libraries are docked with AutoDock; poses are filtered on hinge contact and pocket occupancy rather than raw score, and survivors are re-evaluated with Boltz-2 structure prediction as an orthogonal check. The screen deliberately extends to compounds likely to hold against common resistance mutations. Developed with faculty mentorship at Emory University.
EGFR L858RNSCLCAutoDockBoltz-2drug repurposing
2026 — presentIndependent · USC mentorshipLead researcher
Structure-based repurposing against the dark kinase LMTK3
Lemur tyrosine kinase 3 (LMTK3) is a largely uncharacterized "dark" kinase implicated in endocrine-resistant ER+ breast cancer and, through its regulation of KIT, in gastrointestinal stromal tumors — compelling biology paired with only one prior low-profile computational study and no experimentally validated repurposing screen. This project screens FDA-approved compounds for LMTK3 binding using a curated library of 2,571 approved small molecules (DrugBank merged with current ChEMBL, desalted, property-filtered, and deduplicated). Because the available crystal structure (PDB 6SEQ) captures an inactive DFG-out conformation, the screen is scoped explicitly for type-II binding, and benchmark inhibitors are matched to that conformational state rather than assumed — a validity pitfall the design is built to avoid. Predicted binders are prioritized for biophysical confirmation by MST and ITC, carried out by collaborators at the University of Southern California (Dr. Heinz-Josef Lenz and Dr. Shivani Soni).
LMTK3 · 6SEQER+ breast cancerdrug repurposingtype-IIMST / ITC
May 2026 — presentSeven Lakes HS Cancer Research SocietyLead researcher
Structure-based repurposing against the cancer kinase DCLK1
DCLK1 (doublecortin-like kinase 1) marks tumor stem cells and drives growth, invasion, and chemoresistance in gastrointestinal cancers — strong biology paired with a nearly empty chemical-matter pipeline, since almost no drug-like selective inhibitors exist for it. This study applies the pipeline validated on EGFR to the DCLK1 kinase domain (PDB 7F3G, the ruxolitinib-bound structure), screening an approved-drug library with AutoDock Vina. Score alone is treated as insufficient: poses are filtered for a hinge hydrogen bond at the Val468 backbone, checked against the Met465 gatekeeper, and benchmarked against known binders (ruxolitinib, DCLK1-IN-1) to establish enrichment before any hit is called. Survivors are re-scored with Boltz-2 co-folding to flag structurally implausible poses, then screened for ADMET. Methods and pose-quality filters are pre-registered before the full screen runs, so hit criteria cannot shift after results are seen. Student-led, directed with co-researchers Ashley Jung and Aagrah Singh, whom I trained from no prior background through an eight-module curriculum on kinase structure, docking, and ChimeraX.
DCLK1 · 7F3GAutoDock Vinavirtual screeninghinge H-bond filter
Jun 2026 — presentUT MD Anderson Cancer CenterFirst author (anticipated)
FOLR1 interpretation: biopsy vs. resection concordance
Folate receptor alpha (FRα), encoded by FOLR1, is the companion-diagnostic biomarker gating eligibility for mirvetuximab soravtansine (Elahere) — the first FRα-targeted antibody–drug conjugate approved for platinum-resistant epithelial ovarian, fallopian tube, and primary peritoneal cancer. Positivity requires ≥75% of viable tumor cells at moderate-to-strong (2–3+) membrane staining on the VENTANA FOLR1-2.1 assay, so a single interpretive call at a fixed threshold decides whether a patient receives the therapy — and the sources of variability in that call remain incompletely characterized. My arm of a four-part study quantifies scoring concordance between biopsy and matched surgical resection specimens from the same patients: eligibility is often set on limited biopsy tissue while the resection represents the fuller tumor, so disagreement carries direct treatment consequences. Each case is read in paired analog and whole-slide digital formats with a two-week washout to limit recall bias, and agreement is measured by weighted kappa across specimen type and reading modality. With co-investigators Pablo Mojica, Maximus Montano, and Jacqueline Little; mentored by Dr. Nadia Hameed and Dr. Barrett Lawson. Targeted for AACR and ASCO GU.
FOLR1 / FRαVENTANA FOLR1-2.1weighted kappadigital pathologycompanion Dx
Jun 2026 — presentBaylor College of MedicineMiddle author (anticipated) · ~2-yr horizon
Machine-learning prediction of selective PGK2 small-molecule binders
Phosphoglycerate kinase 2 (PGK2) is the testis-specific isoenzyme of the glycolytic kinase PGK, and selectively inhibiting it over its somatic homolog PGK1 has emerged as a leading route to non-hormonal male contraception by targeting sperm-specific bioenergetics. It is a hard design problem: the two isoforms share a nearly fully conserved binding pocket, differing at only a handful of edge residues, so the pipeline is built to learn selectivity-driving features rather than raw affinity. The workflow integrates DNA-encoded library (DEL) screening signal with structure-based modeling — conformer-ensemble generation, physics-based docking with idock against paired PGK2 and PGK1 structures, ChimeraX pose inspection at the divergent residues, and cross-validation against independent AI scoring and co-folding models to build consensus across orthogonal methods. Survivors are filtered for predicted PGK2-over-PGK1 selectivity and drug-likeness toward an experimentally testable shortlist. A long-horizon project, with publication anticipated once downstream wet-lab confirmation is complete. Conducted in Dr. Damian Young's lab under Dr. Pierce Andrew Jamieson at the Center for Drug Discovery, Baylor College of Medicine.
PGK2 vs PGK1idock / VinaDELChimeraXselectivity
2026 — presentIndependent meta-researchLead · Registered Report
Reporting-quality audit of structure-based virtual screening in oncology
Structure-based virtual screening has become routine in early oncology drug discovery, but its methodological reporting is uneven — validation, benchmarking, and pose-quality criteria are frequently incomplete or defined only after results are known. This meta-research study systematically audits reporting quality across published screens for a pooled panel of roughly ten cancer-associated kinases (including PIM1, WEE1, TTK/MPS1, MELK, MERTK, and HPK1, with DCLK1 as a continuity anchor), scored against a pre-specified rubric so the assessment itself is reproducible. It is being prepared as a Registered Report for a peer-reviewed venue, with co-authors Ashley Jung and Aagrah Singh; a senior author is being recruited before registration.
meta-researchregistered reportreproducibilitykinase docking