Loading…
Loading…
Work
Every project here is an R&D demonstration on public datasets, not client-confidential work. Open one for the method, the data source, and every metric’s caveats.
R&D demonstration, public data · Mar 2026
Ranking candidate kinase inhibitors by predicted binding affinity, ahead of assay screening, using a GNN trained on public ChEMBL data.
0.34 → 0.21
RMSE vs. baseline docking score
TODO-VERIFY — no published held-out split, baseline definition, or seed for this figure yet, and it isn't linked to a notebook or repo. Do not treat as a validated result until it is (CLAUDE.md §9/§10).
R&D demonstration, public data · Mar 2026
Expanding a small fragment library into novel, synthesizable candidates using a diffusion model trained on public ZINC20 fragment data.
61%
Generated molecules passing drug-likeness filters
TODO-VERIFY — the specific QED/synthetic-accessibility thresholds and evaluation set aren't published alongside this figure yet, and it isn't linked to a notebook or repo (CLAUDE.md §9/§10).
R&D demonstration, public data · Apr 2026
Rebuilding a whole-exome variant calling workflow as a containerized Nextflow pipeline, benchmarked on public 1000 Genomes data.
−42%
Runtime and compute cost vs. legacy pipeline
TODO-VERIFY — the 'legacy pipeline' baseline this is measured against isn't documented or published, and there's no linked benchmark script or repo yet (CLAUDE.md §9/§10).
R&D demonstration, public data · Apr 2026
Identifying high-confidence gene targets from a genome-wide CRISPR knockout screen using MAGeCK hit calling on public DepMap data.
18
High-confidence hits shortlisted from 19,000+ genes
TODO-VERIFY — the significance threshold and QC filters behind this shortlist aren't published alongside it yet, and it isn't linked to a notebook or repo (CLAUDE.md §9/§10).
R&D demonstration, public data · May 2026
Annotating cell types across a public PBMC dataset by fine-tuning a scGPT-style single-cell foundation model.
94.1%
Agreement with expert-curated cell labels
TODO-VERIFY — the held-out set and the expert-curated labels it's compared against aren't published alongside this figure yet, and it isn't linked to a notebook or repo (CLAUDE.md §9/§10).
R&D demonstration, public data · May 2026
Localizing immune and stromal cell populations in tumor tissue using spatial deconvolution on a public 10x Visium dataset.
6
Distinct spatial niches resolved within the tumor margin
TODO-VERIFY — the clustering resolution and validation approach behind this niche count aren't published alongside it yet, and it isn't linked to a notebook or repo (CLAUDE.md §9/§10).