Skip to content

Navigation Menu

Sign in
Sign up
@j-fenn
j-fenn
Follow
M.S. Student in AICD3 @ UCSF. ML Bioengineer @ Varosync. ML for drug discovery: model evaluation, benchmarking, perturbation modeling, multi-omics pipelines

Block or report j-fenn

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
j-fenn /README.md

Jack Fenn

M.S. in AI & Computational Drug Discovery and Development, UCSF. Biochemist by training, now working on evaluation and benchmarking for ML in drug discovery.

Currently ML Bioengineer at Varosync. Previously Shoichet Lab and Bivona Lab at UCSF.


Repositories

boltz2-sea-offtarget · Re-analysis of a Boltz-2 vs SEA off-target benchmark, rebuilt from the Lounkine 2012 supplementary data. Both structural confidence metrics — the composite score and ipTM — separate confirmed binders from confirmed non-binders at chance (within-target AUC 0.527 and 0.486), while the affinity outputs reach 0.68–0.71. Also documents a back-transform inconsistency that reversed a reported conclusion, and an audit finding 18% of model inputs carried expression-construct scaffold. Python · RDKit · scikit-learn

perturbation-readouts · GEARS reproduction on Norman/Adamson/Dixit with every common readout computed on identical predictions. Choosing between two MSEs moves the number by ×ばつ; pairing them across models manufactures a ×ばつ "improvement" between models within 4–37% of each other. Direction accuracy sits at 0.19 across all genes and climbs to 0.77 once genes without a measurable response are excluded, 91–96% of true deltas are within 1.96 SE of zero. PyTorch · PyTorch Geometric · scanpy

stk11-3d-remodeling · Hi-C, ATAC and ChIP pipeline for 3D genome remodeling under KRAS-inhibitor resistance across STK11 contexts. Differential contact analysis without biological replicates, a scored ranking rather than manufactured p-values, plus seven pretrained sequence-to-function models and the two silent failure modes (assembly mismatch, chimeric read loss) that nearly ruined it. Ships no study data; runs end to end on a synthetic fixture. Python · cooler · cooltools · pairtools · Nextflow


What I work on

Model evaluation and benchmarking · perturbation modelling · structure and sequence-to-function models · multi-omics pipelines · pharmacovigilance data engineering

Python PyTorch PyTorch Geometric scanpy RDKit cooler/cooltools Nextflow Docker AWS SQL R


San Francisco · LinkedIn · jack.fenn@ucsf.edu

Pinned Loading

  1. boltz2-sea-offtarget boltz2-sea-offtarget Public

    Re-analysis of a Boltz-2 vs SEA off-target benchmark: both structural confidence metrics discriminate at chance, the affinity head does not.

    Jupyter Notebook

  2. perturbation-readouts perturbation-readouts Public

    GEARS reproduction on three public CRISPR perturbation datasets, with every common readout computed on identical predictions.

    Python

  3. stk11-3d-remodeling stk11-3d-remodeling Public

    Hi-C/ATAC/ChIP pipeline for 3D genome remodeling under KRAS-inhibitor resistance. Differential analysis without replicates; runs on a synthetic fixture.

    Python

  4. solvAI solvAI Public

    Forked from ravishankaryadav1/solvAI

    Prototype for an intelligent solvent recommendation platform that helps chemists instantly identify the best solvent for any reaction. Created as part of Next.Sequence Hackathon 2026.

    Jupyter Notebook

AltStyle によって変換されたページ (->オリジナル) /