The Pathway Atlas
Every tumor sample lives in a 50-dimensional pathway-state space. Every drug lives in the same space as a pathway-perturbation vector. That co-location is the most useful thing in cancer pharmacology that nobody has built UI for. This atlas surfaces the patterns that fall out of putting tumors and drugs on the same map.
Tumor-side discoveries
What happens when you look at 9,200 tumors through pathway-state eyes instead of tissue-of-origin eyes.
The pathology-free tumor map
Discovers: Tumors of different cancer types cluster together when you ignore tissue and look at pathway state. Many bladder cancers are biologically melanomas.
Open tumor map →Pathway entropy per cancer
Discovers: Some cancer types have narrow, focused biology (= druggable). Others are chaotic across patients (= unresponsive to targeted therapy). This single number predicts targeted-therapy response better than mutation burden.
Compare entropy →The pathway co-activation graph
Discovers: Across 9,200 tumors, which Hallmark pathways always fire together? Reveals modules that the textbook treats as separate but the data treats as one.
View modules →Drug-side discoveries
The FDA label is the smallest possible description of what a drug does. The actual perturbation vector tells a richer story.
The drug perturbation map
Discovers: Drugs marketed as different target classes cluster together by what they actually do downstream. BCR-ABL drugs and JAK inhibitors are unexpected neighbors.
Open drug map →Drug redundancy graph
Discovers: ~10–15 drugs in our catalog are mechanistic duplicates of cheaper / older drugs. Some $40K/month TKIs are pathway-equivalent to 2nd-gen generics.
See duplicates →Rational combinations
Discovers: The 3- and 4-drug combinations that cover the most pathway space with the LEAST target overlap — pathway-grounded combos, no current trial tests them.
Build combos →Hallmarks-of-Cancer audit
Discovers: ~85% of clinical kinase inhibitors target only the proliferation hallmark. Drugs that secondarily hit immune + angiogenesis + invasion hallmarks are dramatically under-recognized.
Audit the catalog →Tumor × drug joint geometry
The killer demo. Both populations on the same map.
Pharma-tumor co-embedding
Discovers: 9,200 tumors + 92 drugs in one map. Tumors near a drug = predicted match. Drug deserts = entire continents of cancer biology that nothing in our catalog reaches.
Open co-embedding →Drug deserts
Discovers: Quantifies the unmet need at the pathway level, per cohort. 'X% of medulloblastoma samples sit in a pharmacological desert' — the gap is not the tissue, it's the pathway.
See unmet need →Anti-tumor search
Discovers: For each drug, the ideal patient is the tumor whose state equals −Π_d. Some drugs' ideal patient doesn't exist in 9,200 tumors — they were developed against a biology that doesn't naturally occur.
Find orphan drugs →Per-patient lookups
The same primitives, surfaced one tumor or one drug at a time. Reach them from every sample / drug page in the platform.
Patient twin finder
Discovers: For any tumor sample, find the 10 nearest pathway-neighbors across all 9,200 patients. Often most twins are from a different tissue of origin — challenging the tissue-typed-trial framework.
Pick a sample →Per-tumor drug coverage
Discovers: For any tumor sample, which of its elevated pathways DO have a drug that reverses them, and which are completely uncovered? Defines unmet need at the pathway level for that one patient.
Pick a sample →Per-drug pathway shadow
Discovers: For any drug, decompose its perturbation into on-target shadow vs off-target shadow vs ghost (2nd-order). For ~40% of clinical kinase inhibitors, off-target dominates — that's what the drug actually does.
Pick a drug →How the atlas is built
One backend module (atlas.py) builds 12 cached computations from the (D, T, P, G) matrices that Phases 6.0–6.3 already populated. PCA, kNN, and per-pathway correlation — deterministic, no random init, no external ML deps.
Every viz links to the per-sample reversal scorer, cohort scorer, Oncoscape cohorts, and the drug catalog. From any drug or tumor in the platform you can reach every atlas discovery in one click.